)]}'
{"/PATCHSET_LEVEL":[{"author":{"_account_id":38081,"name":"Anthony Galica","display_name":"agalica","email":"anthony.galica@hitachivantara.com","username":"agalica","status":"Hitachi Vantara"},"change_message_id":"8bc11677ee22e3811af674ef543eb683316bc3d8","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":4,"id":"6f19e7f8_311aa200","updated":"2026-06-04 20:13:04.000000000","message":"I think this is a good idea without diving too deep on it at this time.\n\nThe -1 is because the blueprint link does not work.","commit_id":"ac0429e5f11089a6d2f21065cab7c43aadef9238"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"fabc74e366e1c9a15386dbdc5fad1df3f32ba1a8","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":4,"id":"5562e7c8_83dd7479","updated":"2026-06-06 01:05:07.000000000","message":"recheck","commit_id":"ac0429e5f11089a6d2f21065cab7c43aadef9238"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"e95e23af89503f3dd55efe354b1e2fcfe2e4f146","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":4,"id":"266ac85a_d9ce509e","in_reply_to":"6f19e7f8_311aa200","updated":"2026-06-05 00:26:03.000000000","message":"Thanks for the review. I created the Launchpad blueprint and confirmed that the link now resolves:\n\nhttps://blueprints.launchpad.net/cinder/+spec/image-volume-cache-management\n\nI also added the spec and implementation review links to the blueprint whiteboard.","commit_id":"ac0429e5f11089a6d2f21065cab7c43aadef9238"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"22631a88b234b76be87176e410b858c03fa01386","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":5,"id":"2e407c5c_c10ba9c3","updated":"2026-07-16 08:19:25.000000000","message":"@anthony.galica@hitachivantara.com","commit_id":"81b565d1d05903bfcd24b9856f64f233e87c5a64"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"a49783fff1a5c31e5f690a406ac1951f3004377d","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":5,"id":"bfa6ef54_30812460","updated":"2026-07-16 08:21:15.000000000","message":"@anthony.galica@hitachivantara.com\nHi Anthony, could I ask for your review on this spec when you have time?\n\nThe paired implementation is:\nhttps://review.opendev.org/c/openstack/cinder/+/986740\n\nBoth changes are passing Zuul. If you also have bandwidth to look at the implementation, I would appreciate that too. Thanks!","commit_id":"81b565d1d05903bfcd24b9856f64f233e87c5a64"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"4d6e9e7ca7127d90c764a3ecfd09bcf04edb5a7b","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":5,"id":"fc1aa02e_279e6635","updated":"2026-07-27 01:17:14.000000000","message":"Hi @rosmaita.fossdev@gmail.com could I ask for some guidance on this spec when you have bandwidth?\n\nThe paired implementation is:\nhttps://review.opendev.org/c/openstack/cinder/+/986740\n\nBoth changes are passing Zuul and are listed on the Hibiscus review request pad, but the current revisions have not yet received review. I would especially appreciate feedback on whether cinder-manage is the right operator-facing interface, and who might be an appropriate reviewer for the implementation.\n\nThanks!","commit_id":"81b565d1d05903bfcd24b9856f64f233e87c5a64"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"4420af080b88c7ab316c8ab269a791213d2bece8","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":6,"id":"f38a0297_42421480","updated":"2026-10-06 07:12:21.000000000","message":"Hi @anthony.galica@hitachivantara.com would you be able to review this spec, or suggest someone who could?\n\nThe spec has been retargeted to 2027.1, and the paired implementation is passing Zuul:\nhttps://review.opendev.org/c/openstack/cinder/+/986740\n\nI\u0027d especially appreciate initial feedback on the proposed cinder-manage interface. Thanks!","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":6,"id":"702f9ede_8c7c10c9","updated":"2026-10-30 13:53:38.000000000","message":"needs some work","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":7,"id":"93d1a0c3_faec3e07","updated":"2026-11-01 09:55:54.000000000","message":"Thanks for the thorough review, @simon@everpuredata.com. PS7 addresses all seven comments. Proposed change is now split into Target pool, RPC, Output and exit codes, Handler and list. Line numbers in the inline replies refer to PS7. I also dropped the backend size-drift motivation so the spec is about managing the cache itself.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":7,"id":"44293135_f2805ecd","updated":"2026-11-01 17:40:12.000000000","message":"Two Handler rules keep me at -1 for one more round, both one-sentence fixes: the sizing rule at 208 is undefined when Glance reports no `virtual_size` and misses the image-size floor a user create gets, and the pool check at 194 can never pass for the many drivers that report no `pools` list. The rest are clarifications. One request outside the spec: 986740 PS11 still implements the PS6 design (cast, token over RPC, `--host` as a backend, `return 2`), so please rework it to PS7 before anyone reads it as the reference implementation, and drop the Related-Bug and size-drift sentence there as you did here.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":8,"id":"53877e22_a00af61d","updated":"2026-11-02 12:38:15.000000000","message":"@simon@everpuredata.com Thanks for the thorough review. PS8 addresses it.\n\nThe two blocking points: the sizing step (246-261) now uses the floor a user create gets, with virtual_size optional, and the pool check (224-228) follows the scheduler\u0027s naming rule (pool_name, else volume_backend_name, else _pool0). For 148 and 166, a result names the entry\u0027s cluster_name in a cluster, so it matches --host, and a failed line carries entry\u003d when the result has an entry_id. The other comments are addressed as suggested; the replies explain where I went a\ndifferent way (208, 246, 261, 357).\n\nWhile checking the handler against the code I also corrected a few points of the described behaviour: the lookup after the create re-reads the image metadata with a fresh [glance] token, since the create flow reads it itself and the entry carries its updated_at (309-315); validate_host_capacity() accounts for the size in the scheduler\u0027s in-memory pool state, now stated under Performance Impact; and the recovery notes cover a cache volume left without an entry (344-349).\n\n986740 PS12 implements this revision: warm uses a bounded RPC call with no token in the context sent from cinder-manage, --host identifies a pool, and warm sets failure and usage-error exits with sys.exit(). I also removed the Related-Bug trailer and the size-drift sentence from the commit message.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"748f7a21eabc53202777ca4e00c99d2f3383aae7","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":8,"id":"b609b9eb_8d959aa2","updated":"2026-11-10 00:59:12.000000000","message":"Hi @simon@everpuredata.com, This is gentle ping.\nPS8 addresses all your PS7 comments (replied inline). \nCould you take another look before spec freeze on Nov 20? Thanks!","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":8,"id":"cf8b5468_c1479e73","updated":"2026-11-10 19:36:25.000000000","message":"What is left is six minors and three nits, each a sentence or two of spec text; the two that add causes to the no-entry reason (316, 493) also want the string in 986740 updated. +1 now, and I will +2 the patch set that folds them in; the freeze is the 20th, so there is time. 986740 PS13 matches PS8 step for step, thanks for reworking it.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":9,"id":"bf818b7a_eb3ba6c0","updated":"2026-11-11 06:32:35.000000000","message":"@simon@everpuredata.com Thanks for the review. PS9 addresses all nine comments. The related code changes are in 986740 PS15.","commit_id":"182bf4a08c598123ebbc6fc58e11c652f25dd3af"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"48b22fffb241603f13512cac57bfba04e8b5a775","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":9,"id":"f1c2bd27_bf206414","updated":"2026-11-18 10:05:54.000000000","message":"Hi @simon@everpuredata.com, @erlon.rodrigues.cruz@canonical.com, thank you both for the reviews. If you think it\u0027s ready, would one of you mind giving it Workflow +1 before the spec freeze?","commit_id":"182bf4a08c598123ebbc6fc58e11c652f25dd3af"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"76cdee030e16488046041e46553f2b8b36feb58e","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":9,"id":"9997c150_56effe0c","updated":"2026-11-11 14:50:42.000000000","message":"Looks good - all my concerns are addressed","commit_id":"182bf4a08c598123ebbc6fc58e11c652f25dd3af"},{"author":{"_account_id":10058,"name":"Erlon R. Cruz","email":"erlon.rodrigues.cruz@canonical.com","username":"sombrafam"},"change_message_id":"c5fcce8f506a3c467da16c2811b8d3b191908920","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":9,"id":"3ab1963c_ac0e19a7","updated":"2026-11-11 19:31:57.000000000","message":"Looks good to me. It\u0027s a really good idea and will definitely add value to Cinder. Leaving it not +W, case other cores have the chance to look and +W.","commit_id":"182bf4a08c598123ebbc6fc58e11c652f25dd3af"}],"specs/2027.1/image-volume-cache-management.rst":[{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":true,"context_lines":[{"line_number":93,"context_line":"    cinder-manage image-cache list [--image IMAGE_ID]"},{"line_number":94,"context_line":"                                   [--host HOST]"},{"line_number":95,"context_line":""},{"line_number":96,"context_line":"``warm`` resolves the set of target ``cinder-volume`` services and"},{"line_number":97,"context_line":"casts ``warm_image_cache`` to each. The RPC is on the existing"},{"line_number":98,"context_line":"``cinder-volume`` topic at a new version ``3.21`` with"},{"line_number":99,"context_line":"``@rpc.assert_min_rpc_version``, so a mixed-version cluster refuses"}],"source_content_type":"text/x-rst","patch_set":6,"id":"c7be5aa3_2bb193c0","line":96,"updated":"2026-10-30 13:53:38.000000000","message":"`warm` is a cast, so the CLI gets no result and exits 0 even on failure. That defeats the pipeline use case. Use a bounded call, or specify a way to observe the outcome. Give the RPC signature.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[{"line_number":93,"context_line":"    cinder-manage image-cache list [--image IMAGE_ID]"},{"line_number":94,"context_line":"                                   [--host HOST]"},{"line_number":95,"context_line":""},{"line_number":96,"context_line":"``warm`` resolves the set of target ``cinder-volume`` services and"},{"line_number":97,"context_line":"casts ``warm_image_cache`` to each. The RPC is on the existing"},{"line_number":98,"context_line":"``cinder-volume`` topic at a new version ``3.21`` with"},{"line_number":99,"context_line":"``@rpc.assert_min_rpc_version``, so a mixed-version cluster refuses"}],"source_content_type":"text/x-rst","patch_set":6,"id":"e44f41f3_01c35103","line":96,"in_reply_to":"c7be5aa3_2bb193c0","updated":"2026-11-01 09:55:54.000000000","message":"Done. warm is now a call bounded by --timeout (default 600 seconds). It prints created or exists on success (exit 0), or failed, including on an RPC timeout (exit 1); usage errors exit 2 without a result line. The RPC signature is VolumeAPI.warm_image_cache(ctxt, image_id, host, volume_type_id\u003dNone, timeout\u003dNone), guarded by assert_min_rpc_version(\u00273.21\u0027).","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":true,"context_lines":[{"line_number":113,"context_line":"   pool-qualified host can accept it. Size is"},{"line_number":114,"context_line":"   ``max(image.virtual_size_GiB, image.min_disk, 1)`` to satisfy"},{"line_number":115,"context_line":"   ``create_volume_from_image()``."},{"line_number":116,"context_line":"#. Calls ``self.create_volume(internal_ctx, transient, ...,"},{"line_number":117,"context_line":"   allow_reschedule\u003dFalse)`` -- the same path user creates take, so"},{"line_number":118,"context_line":"   the cache is populated by the existing"},{"line_number":119,"context_line":"   ``_create_image_cache_volume_entry()`` side effect."}],"source_content_type":"text/x-rst","patch_set":6,"id":"13e966fd_a9fb512f","line":116,"updated":"2026-10-30 13:53:38.000000000","message":"Several paths leave no cache entry: `clone_image()` returning cloned, the `cinder://` shortcut, encrypted types, and failures in `ensure_space` or the clone. Re-check with `get_entry()` and exit non-zero if there is no entry. Spell out how the handler builds the `request_spec` and the volume.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[{"line_number":113,"context_line":"   pool-qualified host can accept it. Size is"},{"line_number":114,"context_line":"   ``max(image.virtual_size_GiB, image.min_disk, 1)`` to satisfy"},{"line_number":115,"context_line":"   ``create_volume_from_image()``."},{"line_number":116,"context_line":"#. Calls ``self.create_volume(internal_ctx, transient, ...,"},{"line_number":117,"context_line":"   allow_reschedule\u003dFalse)`` -- the same path user creates take, so"},{"line_number":118,"context_line":"   the cache is populated by the existing"},{"line_number":119,"context_line":"   ``_create_image_cache_volume_entry()`` side effect."}],"source_content_type":"text/x-rst","patch_set":6,"id":"2a0bfcb1_80ff3aef","line":116,"in_reply_to":"13e966fd_a9fb512f","updated":"2026-11-01 09:55:54.000000000","message":"Done. After the create, the handler looks the entry up again with get_entry() and returns failed (exit 1) when there is none; lines 246-254 list the paths you mention (clone_image(), cinder://, ensure_space(), a failed clone). Encrypted types are rejected before anything is created (lines 204-207). Lines 225-240 spell out the transient volume record and the request_spec.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":true,"context_lines":[{"line_number":120,"context_line":"#. Deletes the transient in ``finally``. The clone owned by the"},{"line_number":121,"context_line":"   internal tenant remains."},{"line_number":122,"context_line":""},{"line_number":123,"context_line":"Target host resolution for ``warm``:"},{"line_number":124,"context_line":""},{"line_number":125,"context_line":"* ``--host`` given       -\u003e only that host."},{"line_number":126,"context_line":"* ``--volume-type`` given -\u003e all up cinder-volume services whose"}],"source_content_type":"text/x-rst","patch_set":6,"id":"ade0f110_25908d99","line":123,"updated":"2026-10-30 13:53:38.000000000","message":"Which pools are warmed? Multi-pool backends only get one pool, and `clone_across_pools` is not addressed. Note that the create flow\u0027s `image_id` lock makes concurrent casts safe, and say whether `warm` runs once per cluster.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[{"line_number":120,"context_line":"#. Deletes the transient in ``finally``. The clone owned by the"},{"line_number":121,"context_line":"   internal tenant remains."},{"line_number":122,"context_line":""},{"line_number":123,"context_line":"Target host resolution for ``warm``:"},{"line_number":124,"context_line":""},{"line_number":125,"context_line":"* ``--host`` given       -\u003e only that host."},{"line_number":126,"context_line":"* ``--volume-type`` given -\u003e all up cinder-volume services whose"}],"source_content_type":"text/x-rst","patch_set":6,"id":"f9146bcb_c002f213","line":123,"in_reply_to":"ade0f110_25908d99","updated":"2026-11-01 09:55:54.000000000","message":"Done. Each invocation now targets one pool, given as --host host@backend#pool (or cluster@backend#pool), so a multi-pool backend is warmed with one run per pool (lines 92-107). The lookup follows the create flow, including clone_across_pools (lines 197-203). Concurrent runs rely on the create flow\u0027s image-id lock: one entry is recorded and each run reports it as created or exists (lines 261-268). In a cluster the call goes to the cluster and the entry is keyed by cluster@backend#pool, so one run per pool of the cluster is enough (lines 113-118).","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":true,"context_lines":[{"line_number":123,"context_line":"Target host resolution for ``warm``:"},{"line_number":124,"context_line":""},{"line_number":125,"context_line":"* ``--host`` given       -\u003e only that host."},{"line_number":126,"context_line":"* ``--volume-type`` given -\u003e all up cinder-volume services whose"},{"line_number":127,"context_line":"  ``host`` ends in the type\u0027s ``volume_backend_name``."},{"line_number":128,"context_line":"* neither given          -\u003e the same lookup using"},{"line_number":129,"context_line":"  ``CONF.default_volume_type``. If unset, ``warm`` exits 2 and asks"}],"source_content_type":"text/x-rst","patch_set":6,"id":"a7a351b6_0641332f","line":126,"updated":"2026-10-30 13:53:38.000000000","message":"\"Host ends in `volume_backend_name`\" is wrong. A service host is `node@\u003csection\u003e[#pool]`, and `volume_backend_name` is a separate capability. Many types have no such extra spec. Say how the mapping is done (the scheduler backend state), what happens when the type has none, and what format `--host` takes.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[{"line_number":123,"context_line":"Target host resolution for ``warm``:"},{"line_number":124,"context_line":""},{"line_number":125,"context_line":"* ``--host`` given       -\u003e only that host."},{"line_number":126,"context_line":"* ``--volume-type`` given -\u003e all up cinder-volume services whose"},{"line_number":127,"context_line":"  ``host`` ends in the type\u0027s ``volume_backend_name``."},{"line_number":128,"context_line":"* neither given          -\u003e the same lookup using"},{"line_number":129,"context_line":"  ``CONF.default_volume_type``. If unset, ``warm`` exits 2 and asks"}],"source_content_type":"text/x-rst","patch_set":6,"id":"d6314529_402bf130","line":126,"in_reply_to":"a7a351b6_0641332f","updated":"2026-11-01 09:55:54.000000000","message":"Agreed, that sentence was wrong and is gone. warm no longer maps volume types to hosts: the pool is named explicitly by --host, in the format openstack volume backend pool list shows (lines 92-96). --volume-type now only sets the type of the transient (and so the cache) volume, which changes its meaning from PS6 (lines 109-111). Before creating anything, the handler asks the scheduler, via validate_host_capacity(), whether that type passes the filters for the pool; a type without volume_backend_name is judged by its other extra specs, as in a normal create (lines 212-221).","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":true,"context_lines":[{"line_number":125,"context_line":"* ``--host`` given       -\u003e only that host."},{"line_number":126,"context_line":"* ``--volume-type`` given -\u003e all up cinder-volume services whose"},{"line_number":127,"context_line":"  ``host`` ends in the type\u0027s ``volume_backend_name``."},{"line_number":128,"context_line":"* neither given          -\u003e the same lookup using"},{"line_number":129,"context_line":"  ``CONF.default_volume_type``. If unset, ``warm`` exits 2 and asks"},{"line_number":130,"context_line":"  for ``--host`` or ``--volume-type``."},{"line_number":131,"context_line":""}],"source_content_type":"text/x-rst","patch_set":6,"id":"6903b267_83bb6b85","line":128,"updated":"2026-10-30 13:53:38.000000000","message":"Default-type resolution is described inconsistently: the CLI resolves it at 128-130, but a manager branch at 334 also handles it. The internal tenant is a hard prerequisite, so list it under deployer impact. The quota remark should count two volumes.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[{"line_number":125,"context_line":"* ``--host`` given       -\u003e only that host."},{"line_number":126,"context_line":"* ``--volume-type`` given -\u003e all up cinder-volume services whose"},{"line_number":127,"context_line":"  ``host`` ends in the type\u0027s ``volume_backend_name``."},{"line_number":128,"context_line":"* neither given          -\u003e the same lookup using"},{"line_number":129,"context_line":"  ``CONF.default_volume_type``. If unset, ``warm`` exits 2 and asks"},{"line_number":130,"context_line":"  for ``--host`` or ``--volume-type``."},{"line_number":131,"context_line":""}],"source_content_type":"text/x-rst","patch_set":6,"id":"9df0ca14_6e47d92d","line":128,"in_reply_to":"6903b267_83bb6b85","updated":"2026-11-01 09:55:54.000000000","message":"Done. The default type is resolved only in the handler (lines 204-205); the CLI just resolves a given --volume-type to an ID (lines 136-140). The internal tenant is listed under Other deployer impact as a hard prerequisite (lines 400-402), and the quota note under Performance Impact now counts two volumes (lines 378-382).","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":true,"context_lines":[{"line_number":187,"context_line":"* Token is service-user scoped (typically project ``service``, user"},{"line_number":188,"context_line":"  ``cinder``) -- the same identity cinder-volume already presents to"},{"line_number":189,"context_line":"  Glance for user creates from images."},{"line_number":190,"context_line":"* Token is short-lived, in-process only, never written to disk or"},{"line_number":191,"context_line":"  logged."},{"line_number":192,"context_line":"* ``cinder-manage`` is admin-only by virtue of requiring direct"},{"line_number":193,"context_line":"  access to the cinder config and DB."}],"source_content_type":"text/x-rst","patch_set":6,"id":"5a9a3f16_a77626ab","line":190,"updated":"2026-10-30 13:53:38.000000000","message":"\"In-process only, never logged\" is false. `to_dict()` includes `auth_token`, so the token goes over RPC. Either say so and justify it, or have c-vol get its own token. Also state the role and whether it can read other projects\u0027 private images.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[{"line_number":187,"context_line":"* Token is service-user scoped (typically project ``service``, user"},{"line_number":188,"context_line":"  ``cinder``) -- the same identity cinder-volume already presents to"},{"line_number":189,"context_line":"  Glance for user creates from images."},{"line_number":190,"context_line":"* Token is short-lived, in-process only, never written to disk or"},{"line_number":191,"context_line":"  logged."},{"line_number":192,"context_line":"* ``cinder-manage`` is admin-only by virtue of requiring direct"},{"line_number":193,"context_line":"  access to the cinder config and DB."}],"source_content_type":"text/x-rst","patch_set":6,"id":"5e0a22ff_08215d22","line":190,"in_reply_to":"5a9a3f16_a77626ab","updated":"2026-11-01 09:55:54.000000000","message":"You\u0027re right, the token did go over RPC. I went with the second option: the RPC from cinder-manage now carries no token, and cinder-volume gets its own token with the [glance] service credentials (lines 311-322). The role, and that with the admin role Kolla gives the cinder user it can read other projects\u0027 private images, are covered in lines 323-327; that is why warm is only exposed through cinder-manage.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"c625e679598d36f728548cd78d6122ab3e74f940","unresolved":true,"context_lines":[{"line_number":312,"context_line":"  * Bug #1968164 -- \"Incorrectly shown volume size while using"},{"line_number":313,"context_line":"    PowerFlex\""},{"line_number":314,"context_line":""},{"line_number":315,"context_line":"  This spec does not claim to fix those bugs; it gives operators a"},{"line_number":316,"context_line":"  way to take the affected code path before user-visible volumes are"},{"line_number":317,"context_line":"  created, so the drift problem is not triggered in practice."},{"line_number":318,"context_line":""}],"source_content_type":"text/x-rst","patch_set":6,"id":"75e03018_d038c73e","line":315,"updated":"2026-10-30 13:53:38.000000000","message":"The mitigation only holds until eviction or an image update. Say `warm` must be re-run, and that the real fix belongs in the driver.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"50a38780d3e704e6780b92fafdab0703f2e735ee","unresolved":false,"context_lines":[{"line_number":312,"context_line":"  * Bug #1968164 -- \"Incorrectly shown volume size while using"},{"line_number":313,"context_line":"    PowerFlex\""},{"line_number":314,"context_line":""},{"line_number":315,"context_line":"  This spec does not claim to fix those bugs; it gives operators a"},{"line_number":316,"context_line":"  way to take the affected code path before user-visible volumes are"},{"line_number":317,"context_line":"  created, so the drift problem is not triggered in practice."},{"line_number":318,"context_line":""}],"source_content_type":"text/x-rst","patch_set":6,"id":"caf4a491_2aa44c9c","line":315,"in_reply_to":"75e03018_d038c73e","updated":"2026-11-01 09:55:54.000000000","message":"Lines 415-417 now say a warmed entry lasts until it is evicted or its image is updated, and that warm has to be re-run after that. I removed the driver-bug mitigation from the spec so it stays focused on managing the cache itself, so the driver part no longer applies.","commit_id":"412dc5df534faca5594326037fa4998233ebdf90"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":103,"context_line":"  instead."},{"line_number":104,"context_line":"* If nothing is up for the service or cluster, the pool is reported"},{"line_number":105,"context_line":"  as failed (exit 1) without sending anything."},{"line_number":106,"context_line":"* A backend with several pools is warmed by running ``warm`` once per"},{"line_number":107,"context_line":"  pool."},{"line_number":108,"context_line":""},{"line_number":109,"context_line":"The pool is named explicitly to keep the scope of each run clear;"}],"source_content_type":"text/x-rst","patch_set":7,"id":"1cef1f54_b5cc4edc","line":106,"updated":"2026-11-01 17:40:12.000000000","message":"Still open from PS6 (my L123). Step 6 handles `clone_across_pools` correctly, but this bullet and 117-118 \"each host@backend#pool has its own entries\" were not reconciled with it: `_get_query_filters` returns no filter when not clustered and the driver reports `clone_across_pools` (`image/cache.py:58-70`), so entries are shared across the backend\u0027s pools, and 117-118 is false for such a backend. In-tree only StorPool reports it (`storpool.py:370`); NetApp hard-codes False (`block_cmode.py:93`). Please qualify both: once per pool except for a backend reporting `clone_across_pools`, where one run is enough and runs on its other pools return `exists` naming that entry; and each pool has its own entries unless the driver reports `clone_across_pools`.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":103,"context_line":"  instead."},{"line_number":104,"context_line":"* If nothing is up for the service or cluster, the pool is reported"},{"line_number":105,"context_line":"  as failed (exit 1) without sending anything."},{"line_number":106,"context_line":"* A backend with several pools is warmed by running ``warm`` once per"},{"line_number":107,"context_line":"  pool."},{"line_number":108,"context_line":""},{"line_number":109,"context_line":"The pool is named explicitly to keep the scope of each run clear;"}],"source_content_type":"text/x-rst","patch_set":7,"id":"f1bfc8f9_ad79a00b","line":106,"in_reply_to":"1cef1f54_b5cc4edc","updated":"2026-11-02 12:38:15.000000000","message":"Done at 105-110 and 123-126. See my reply at L261 for the concurrent-run case where the initial lookup misses.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":108,"context_line":""},{"line_number":109,"context_line":"The pool is named explicitly to keep the scope of each run clear;"},{"line_number":110,"context_line":"targets are not resolved from a volume type. ``--volume-type`` only"},{"line_number":111,"context_line":"sets the type of the transient volume, and so of the cache volume."},{"line_number":112,"context_line":""},{"line_number":113,"context_line":"In a cluster, the volumes and cache entries of a pool carry the"},{"line_number":114,"context_line":"``cluster_name`` ``cluster@backend#pool`` and every member looks"}],"source_content_type":"text/x-rst","patch_set":7,"id":"7ed766e2_72b6061e","line":111,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: true only when this run records the entry. The lookup at 197-203 is by `cluster_name` or host alone, so `warm --volume-type gold` on a pool whose entry\u0027s cache volume is silver returns `exists` and retypes nothing, and in the race at 261-268 the run that loses the lock clones from the other run\u0027s entry, whose cache volume carries that run\u0027s type, although it reports `created`. One sentence: an existing entry is used whatever its cache volume\u0027s type; `--volume-type` is not compared with it and never retypes it.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":108,"context_line":""},{"line_number":109,"context_line":"The pool is named explicitly to keep the scope of each run clear;"},{"line_number":110,"context_line":"targets are not resolved from a volume type. ``--volume-type`` only"},{"line_number":111,"context_line":"sets the type of the transient volume, and so of the cache volume."},{"line_number":112,"context_line":""},{"line_number":113,"context_line":"In a cluster, the volumes and cache entries of a pool carry the"},{"line_number":114,"context_line":"``cluster_name`` ``cluster@backend#pool`` and every member looks"}],"source_content_type":"text/x-rst","patch_set":7,"id":"c4823497_ad20998d","line":111,"in_reply_to":"7ed766e2_72b6061e","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":122,"context_line":""},{"line_number":123,"context_line":"``warm`` waits for the result. The RPC is a ``call`` on the existing"},{"line_number":124,"context_line":"``cinder-volume`` topic of the target service or cluster, at a new"},{"line_number":125,"context_line":"version ``3.21``:"},{"line_number":126,"context_line":""},{"line_number":127,"context_line":"::"},{"line_number":128,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"bc09ebdd_6c314ca9","line":125,"updated":"2026-11-01 17:40:12.000000000","message":"Nit, no vote impact: master is 3.20 (`volume/rpcapi.py:145`) and 3.21 is already spoken for twice, by the merged 2027.1 spec `retype-volume-including-snapshots.rst` (398-409, which assigns 3.21 and 3.22) and by cinder 1000900. Whichever lands second renumbers, so the four literals at 125, 144, 413 and 445 are provisional; write \"the next VolumeAPI version (3.21 at the time of writing)\" once here and \"the new version\" elsewhere.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":122,"context_line":""},{"line_number":123,"context_line":"``warm`` waits for the result. The RPC is a ``call`` on the existing"},{"line_number":124,"context_line":"``cinder-volume`` topic of the target service or cluster, at a new"},{"line_number":125,"context_line":"version ``3.21``:"},{"line_number":126,"context_line":""},{"line_number":127,"context_line":"::"},{"line_number":128,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"43afa60b_53f94b00","line":125,"in_reply_to":"bc09ebdd_6c314ca9","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":145,"context_line":"  so while older cinder-volume services remain, it raises"},{"line_number":146,"context_line":"  ``ServiceTooOld`` without sending anything."},{"line_number":147,"context_line":"* It returns ``{\u0027status\u0027: \u0027created\u0027 | \u0027exists\u0027 | \u0027failed\u0027, \u0027host\u0027:"},{"line_number":148,"context_line":"  ..., \u0027entry_id\u0027: ..., \u0027reason\u0027: ...}``. ``host`` is the host of the"},{"line_number":149,"context_line":"  cache entry for ``created`` and ``exists``, and for ``failed`` when"},{"line_number":150,"context_line":"  an entry was recorded but the transient volume was left behind;"},{"line_number":151,"context_line":"  otherwise it is the requested pool."}],"source_content_type":"text/x-rst","patch_set":7,"id":"7882a295_2dc63410","line":148,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: in a cluster the entry\u0027s host column is the handling member\u0027s `member@backend#pool`: step 10 gives the transient the member\u0027s host, `_clone_image_volume` copies host and `cluster_name` into the cache volume (neither is in `_VOLUME_CLONE_SKIP_PROPERTIES`, `manager.py:244-248`) and `create_cache_entry` stores both. So `created` and `exists` return `node3@backend#pool` while the operator passed `cluster@backend#pool`, and `failed` without an entry echoes the cluster string (151). Say which it is, e.g. \"in a cluster, `host` is the entry\u0027s `cluster_name`, so it matches `--host`\", or print both as the `list` table does, and apply the same to 164-166 and 202-203.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":145,"context_line":"  so while older cinder-volume services remain, it raises"},{"line_number":146,"context_line":"  ``ServiceTooOld`` without sending anything."},{"line_number":147,"context_line":"* It returns ``{\u0027status\u0027: \u0027created\u0027 | \u0027exists\u0027 | \u0027failed\u0027, \u0027host\u0027:"},{"line_number":148,"context_line":"  ..., \u0027entry_id\u0027: ..., \u0027reason\u0027: ...}``. ``host`` is the host of the"},{"line_number":149,"context_line":"  cache entry for ``created`` and ``exists``, and for ``failed`` when"},{"line_number":150,"context_line":"  an entry was recorded but the transient volume was left behind;"},{"line_number":151,"context_line":"  otherwise it is the requested pool."}],"source_content_type":"text/x-rst","patch_set":7,"id":"b9f0d204_9f2162f0","line":148,"in_reply_to":"7882a295_2dc63410","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":150,"context_line":"  an entry was recorded but the transient volume was left behind;"},{"line_number":151,"context_line":"  otherwise it is the requested pool."},{"line_number":152,"context_line":""},{"line_number":153,"context_line":"``warm`` disables ``[oslo_messaging_rabbit] use_queue_manager`` for"},{"line_number":154,"context_line":"its own process, so it waits on a randomly named reply queue and"},{"line_number":155,"context_line":"cannot share one with a running service or another ``cinder-manage``."},{"line_number":156,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"89bf8501_d45982fe","line":153,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: the rule is right and necessary, but the sentence gives no mechanism and names an implausible partner. With the queue manager on, the reply queue is named `\u003chostname\u003e:\u003cprocessname\u003e:\u003ccounter\u003e` and `QManager.get()` restarts the counter for each new process group (`oslo.messaging amqpdriver.py:82-113`), so two `cinder-manage` runs started from different shells both wait on `..:cinder-manage:1` and a reply delivered to the wrong one is dropped. A running cinder-volume has a different `processname` (basename of argv[0], `impl_rabbit.py:263-266`) and cannot collide unless `processname` is overridden. Suggest: \"`warm` disables `use_queue_manager` for its own process: with it, reply queue names restart their counter per process group, so two concurrent `cinder-manage` runs on one host could share a queue and lose each other\u0027s replies.\" 986740 does not implement the override yet.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":150,"context_line":"  an entry was recorded but the transient volume was left behind;"},{"line_number":151,"context_line":"  otherwise it is the requested pool."},{"line_number":152,"context_line":""},{"line_number":153,"context_line":"``warm`` disables ``[oslo_messaging_rabbit] use_queue_manager`` for"},{"line_number":154,"context_line":"its own process, so it waits on a randomly named reply queue and"},{"line_number":155,"context_line":"cannot share one with a running service or another ``cinder-manage``."},{"line_number":156,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"20e82c3e_4aaff3e4","line":153,"in_reply_to":"89bf8501_d45982fe","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":163,"context_line":""},{"line_number":164,"context_line":"    created: image\u003d\u003cimage-id\u003e host\u003d\u003centry-host\u003e entry\u003d\u003centry-id\u003e"},{"line_number":165,"context_line":"    exists: image\u003d\u003cimage-id\u003e host\u003d\u003centry-host\u003e entry\u003d\u003centry-id\u003e"},{"line_number":166,"context_line":"    failed: image\u003d\u003cimage-id\u003e host\u003d\u003creported-host\u003e reason\u003d\u003creason\u003e"},{"line_number":167,"context_line":""},{"line_number":168,"context_line":"\u003d\u003d\u003d\u003d  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":169,"context_line":"Exit  Meaning"}],"source_content_type":"text/x-rst","patch_set":7,"id":"edad4232_81dbc81a","line":166,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: 149-150 and 256-257 say the failed result names the entry and the transient when a volume is left behind, but the dictionary at 147-148 has no field for the volume and this line has no `entry\u003d`, so both can only ride inside the free-form reason. Either add an optional `entry\u003d\u003centry-id\u003e` here when `entry_id` is set and say at 257 that `reason` carries the transient\u0027s ID, or drop \"and the cache entry if one was recorded\". Optionally add after 259 that `warm` never deletes a transient left by an earlier run, so the operator removes it like any internal-tenant volume.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":163,"context_line":""},{"line_number":164,"context_line":"    created: image\u003d\u003cimage-id\u003e host\u003d\u003centry-host\u003e entry\u003d\u003centry-id\u003e"},{"line_number":165,"context_line":"    exists: image\u003d\u003cimage-id\u003e host\u003d\u003centry-host\u003e entry\u003d\u003centry-id\u003e"},{"line_number":166,"context_line":"    failed: image\u003d\u003cimage-id\u003e host\u003d\u003creported-host\u003e reason\u003d\u003creason\u003e"},{"line_number":167,"context_line":""},{"line_number":168,"context_line":"\u003d\u003d\u003d\u003d  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":169,"context_line":"Exit  Meaning"}],"source_content_type":"text/x-rst","patch_set":7,"id":"040969da_8b80a3bc","line":166,"in_reply_to":"edad4232_81dbc81a","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":168,"context_line":"\u003d\u003d\u003d\u003d  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":169,"context_line":"Exit  Meaning"},{"line_number":170,"context_line":"\u003d\u003d\u003d\u003d  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":171,"context_line":"0     ``created`` or ``exists``: an entry was found or created under"},{"line_number":172,"context_line":"      the lookup rule a create on the pool uses (see Handler)."},{"line_number":173,"context_line":"1     ``failed``, including no reply within ``--timeout`` (the"},{"line_number":174,"context_line":"      backend may still be working), backends too old for the RPC,"}],"source_content_type":"text/x-rst","patch_set":7,"id":"68410381_6f08b132","line":171,"updated":"2026-11-01 17:40:12.000000000","message":"Still open from PS6 (my L96). The table is the contract I asked for, but the spec does not say how the process status is set, and the in-tree precedent is the broken one: `cinder-manage main()` ends with `fn(**fn_kwargs)` and discards the return value (`cmd/manage.py`, last lines), so only actions that call `sys.exit()` themselves, as the db commands do, exit non-zero; the `return 2` in `service remove` (:782) and the cluster commands (:831, :842, :892) exit 0 today. 986740 copies that pattern (six `return 2`, `return None` at the end of `warm`), so as it stands every outcome would exit 0 and the pipeline use case at 53-55 would not work. One sentence here: warm sets the exit status with `sys.exit()`, as the db commands do, since `main()` ignores an action\u0027s return value; and have the `test_cmd.py` work item assert `SystemExit.code` rather than a return value. Please do not change `main()` in this change, that would alter the three existing commands.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":168,"context_line":"\u003d\u003d\u003d\u003d  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":169,"context_line":"Exit  Meaning"},{"line_number":170,"context_line":"\u003d\u003d\u003d\u003d  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":171,"context_line":"0     ``created`` or ``exists``: an entry was found or created under"},{"line_number":172,"context_line":"      the lookup rule a create on the pool uses (see Handler)."},{"line_number":173,"context_line":"1     ``failed``, including no reply within ``--timeout`` (the"},{"line_number":174,"context_line":"      backend may still be working), backends too old for the RPC,"}],"source_content_type":"text/x-rst","patch_set":7,"id":"5d06d8c9_82507a0c","line":171,"in_reply_to":"68410381_6f08b132","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":190,"context_line":"#. Fails if ``[glance] auth_type`` is not set; otherwise gets a token"},{"line_number":191,"context_line":"   with the ``[glance]`` service credentials and puts it on the"},{"line_number":192,"context_line":"   internal-tenant context (see Security impact)."},{"line_number":193,"context_line":"#. Fetches the image metadata from Glance with that context."},{"line_number":194,"context_line":"#. Checks that ``pool`` is one of the pools the driver reports"},{"line_number":195,"context_line":"   (``driver.get_volume_stats(refresh\u003dFalse)``); otherwise it fails"},{"line_number":196,"context_line":"   and lists them."}],"source_content_type":"text/x-rst","patch_set":7,"id":"d5888737_f0e20579","line":193,"updated":"2026-11-01 17:40:12.000000000","message":"Step 4 is the only Handler step without a \"Fails if\". A user create rejects a non-active image in the API (`volume_utils.check_image_metadata`, `volume_utils.py:1104`), but the manager flow the handler calls has no such check, so for a queued, saving, importing or deactivated image warm reserves quota, creates the transient, fails on the download (`ImageDownloadFailed` \"image contains no data\", `image/glance.py:514-515`, or `ImageNotAuthorized` for a deactivated one), deletes, and reports failed with a download reason instead of \"image not active\". A Packer post-processor makes a not-yet-active image a realistic input, and such an image also has no `virtual_size`, which bears on 208. Please add: fails unless `image.status` is active, with the status in the reason; and say an image Glance does not return (deleted, or not visible to the `[glance]` identity) fails with `ImageNotFound`.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":190,"context_line":"#. Fails if ``[glance] auth_type`` is not set; otherwise gets a token"},{"line_number":191,"context_line":"   with the ``[glance]`` service credentials and puts it on the"},{"line_number":192,"context_line":"   internal-tenant context (see Security impact)."},{"line_number":193,"context_line":"#. Fetches the image metadata from Glance with that context."},{"line_number":194,"context_line":"#. Checks that ``pool`` is one of the pools the driver reports"},{"line_number":195,"context_line":"   (``driver.get_volume_stats(refresh\u003dFalse)``); otherwise it fails"},{"line_number":196,"context_line":"   and lists them."}],"source_content_type":"text/x-rst","patch_set":7,"id":"b43d6122_6d311683","line":193,"in_reply_to":"d5888737_f0e20579","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":191,"context_line":"   with the ``[glance]`` service credentials and puts it on the"},{"line_number":192,"context_line":"   internal-tenant context (see Security impact)."},{"line_number":193,"context_line":"#. Fetches the image metadata from Glance with that context."},{"line_number":194,"context_line":"#. Checks that ``pool`` is one of the pools the driver reports"},{"line_number":195,"context_line":"   (``driver.get_volume_stats(refresh\u003dFalse)``); otherwise it fails"},{"line_number":196,"context_line":"   and lists them."},{"line_number":197,"context_line":"#. Looks the entry up with ``ImageVolumeCache.get_entry()`` the way"}],"source_content_type":"text/x-rst","patch_set":7,"id":"5cdb3144_a45cee02","line":194,"updated":"2026-11-01 17:40:12.000000000","message":"Still open from PS6 (my L123). Many in-tree drivers (RBD, generic NFS, SolidFire and others; Pure reports pools and is unaffected) return stats with no `pools` list, and for them the scheduler names the single pool after `volume_backend_name`, or `_pool0` when that is unset (`scheduler/host_manager.py:252-262`, `volume_utils.py:753`). That is the `#pool` that openstack volume backend pool list prints and that `volume.host` and the cache entry carry, so the name lines 92-96 tell the operator to copy can never be \"one of the pools the driver reports\", and \"lists them\" prints nothing. PS6 had the reconciling sentence at its 111-113 and PS7 dropped it; 986740 already needed a `pools[0]` / `pool_name` / `volume_backend_name` fallback for exactly this. Either define step 5 by the scheduler\u0027s rule (a `pool_name` from `stats[\u0027pools\u0027]`, otherwise the stats\u0027 `volume_backend_name`, otherwise `_pool0`), or drop it and let step 8 carry the check, since `backend_passes_filters` already compares the pool against every `PoolState` and raises `NoValidBackend` for an unknown one; if so, say the failure reason lists the backend\u0027s pools. While here, 132-135 should say the message carries the bare pool name after `#` and the handler rebuilds `host@backend#pool` from its own host, and `cluster@backend#pool` from its cluster, for steps 6, 8 and 10.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":191,"context_line":"   with the ``[glance]`` service credentials and puts it on the"},{"line_number":192,"context_line":"   internal-tenant context (see Security impact)."},{"line_number":193,"context_line":"#. Fetches the image metadata from Glance with that context."},{"line_number":194,"context_line":"#. Checks that ``pool`` is one of the pools the driver reports"},{"line_number":195,"context_line":"   (``driver.get_volume_stats(refresh\u003dFalse)``); otherwise it fails"},{"line_number":196,"context_line":"   and lists them."},{"line_number":197,"context_line":"#. Looks the entry up with ``ImageVolumeCache.get_entry()`` the way"}],"source_content_type":"text/x-rst","patch_set":7,"id":"46335f40_18a1611f","line":194,"in_reply_to":"5cdb3144_a45cee02","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":199,"context_line":"   entry a create on the pool would try to clone. In a cluster the"},{"line_number":200,"context_line":"   lookup is limited to ``cluster@backend#pool``; otherwise to"},{"line_number":201,"context_line":"   ``host@backend#pool``, or not limited by host when the driver"},{"line_number":202,"context_line":"   reports ``clone_across_pools``. A hit returns ``exists`` with the"},{"line_number":203,"context_line":"   entry\u0027s host."},{"line_number":204,"context_line":"#. Picks the volume type: ``volume_type_id`` if given, else"},{"line_number":205,"context_line":"   ``[DEFAULT] default_volume_type`` of this cinder-volume. Fails if"}],"source_content_type":"text/x-rst","patch_set":7,"id":"689a00fc_9c75b796","line":202,"updated":"2026-11-01 17:40:12.000000000","message":"`exists` is returned for whatever `entry get_entry()` finds, and that checks only the image\u0027s `updated_at` (`image/cache.py:259-273`), never the volumes table, so an entry whose cache volume is gone on the backend, or whose row was removed with `cinder-manage volume delete` (which casts `delete_volume` and skips the eviction at `volume/api.py:540`), makes `warm` report `exists` and exit 0 while every user create on the pool keeps downloading. That mirrors the create flow, as 197-199 say, so I am not asking for a check, but the use case at 57-58, \"after repairing its backend\", promises something `warm` cannot deliver. Please reword that example (e.g. \"after its entry was evicted or its cache volume was deleted\") and add after 203 that `warm` trusts the entry as a create does, and that a stale entry is replaced by deleting the cache volume through the volume API, which evicts it, then running `warm` again.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":199,"context_line":"   entry a create on the pool would try to clone. In a cluster the"},{"line_number":200,"context_line":"   lookup is limited to ``cluster@backend#pool``; otherwise to"},{"line_number":201,"context_line":"   ``host@backend#pool``, or not limited by host when the driver"},{"line_number":202,"context_line":"   reports ``clone_across_pools``. A hit returns ``exists`` with the"},{"line_number":203,"context_line":"   entry\u0027s host."},{"line_number":204,"context_line":"#. Picks the volume type: ``volume_type_id`` if given, else"},{"line_number":205,"context_line":"   ``[DEFAULT] default_volume_type`` of this cinder-volume. Fails if"}],"source_content_type":"text/x-rst","patch_set":7,"id":"4ee7cead_2de460f0","line":202,"in_reply_to":"689a00fc_9c75b796","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":205,"context_line":"   ``[DEFAULT] default_volume_type`` of this cinder-volume. Fails if"},{"line_number":206,"context_line":"   the type does not exist, or is encrypted, since the create flow"},{"line_number":207,"context_line":"   never caches encrypted volumes."},{"line_number":208,"context_line":"#. Sizes the transient volume at ``max(ceil(image.virtual_size GiB),"},{"line_number":209,"context_line":"   image.min_disk, 1)``, so that the create flow accepts it"},{"line_number":210,"context_line":"   (``image_utils.check_virtual_size()``) and it meets the image\u0027s"},{"line_number":211,"context_line":"   ``min_disk``."}],"source_content_type":"text/x-rst","patch_set":7,"id":"2051d99c_76838237","line":208,"updated":"2026-11-01 17:40:12.000000000","message":"This formula is undefined when Glance reports no `virtual_size`, which Cinder treats as a normal case (`cinder/volume/api.py:2726-2727` notes an image_snap-backed image has size 0 and `virtual_size` None, and both `volume_utils.py:1129` and `flows/manager/create_volume.py:1096` guard it). 986740 reads it as 0, so the transient becomes `max(min_disk, 1)` GiB, and \"so that the create flow accepts it\" does not hold: the handler calls `VolumeManager.create_volume()` directly and so skips `volume_utils.check_image_metadata()` (`volume_utils.py:1100-1134`, called only from `flows/api/create_volume.py:251` and reimage), and the only size gate on its path is `check_virtual_size()` at `flows/manager/create_volume.py:1010-1011`, which runs after `TemporaryImages.fetch` has downloaded the whole image. A 10 GiB raw image with `min_disk` 0 is downloaded in full and then reported failed, and `warm` has no `--size`. Please use the floor a user create gets: `max(ceil(image.size GiB), ceil(image.virtual_size GiB) when present, image.min_disk, 1)`, say `virtual_size` may be absent, and add to 246-254 that an image whose real virtual size exceeds that floor fails after the download, with setting `min_disk` on the image as the remedy. One precision for 378-381: the flow sets `volume.size` to `ceil(virtual_size)` before the cache clone (`create_volume.py:1013-1016`), so \"twice the size\" is exact only when `min_disk` is at most the virtual size.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":205,"context_line":"   ``[DEFAULT] default_volume_type`` of this cinder-volume. Fails if"},{"line_number":206,"context_line":"   the type does not exist, or is encrypted, since the create flow"},{"line_number":207,"context_line":"   never caches encrypted volumes."},{"line_number":208,"context_line":"#. Sizes the transient volume at ``max(ceil(image.virtual_size GiB),"},{"line_number":209,"context_line":"   image.min_disk, 1)``, so that the create flow accepts it"},{"line_number":210,"context_line":"   (``image_utils.check_virtual_size()``) and it meets the image\u0027s"},{"line_number":211,"context_line":"   ``min_disk``."}],"source_content_type":"text/x-rst","patch_set":7,"id":"d175bd61_df02ba28","line":208,"in_reply_to":"2051d99c_76838237","updated":"2026-11-02 12:38:15.000000000","message":"Done at 246-261 with that floor, with the units spelled out (size and virtual_size in bytes, min_disk in GiB), and a note that Glance may not report virtual_size.\n\nThe note on an actual virtual size above the floor is in the same step rather than at the old 246-254 (now 309-327): that case is a failure of the create itself, raised by check_virtual_size() after the download and reported as \"transient volume creation failed: ...\", not a missing cache entry. The remedy is given as setting min_disk to at least the actual virtual size rounded up to GiB.\n\nThe 378-381 point is covered by the quota bullet (487-494), which now gives the cache volume\u0027s size as the virtual size rounded up to GiB.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":209,"context_line":"   image.min_disk, 1)``, so that the create flow accepts it"},{"line_number":210,"context_line":"   (``image_utils.check_virtual_size()``) and it meets the image\u0027s"},{"line_number":211,"context_line":"   ``min_disk``."},{"line_number":212,"context_line":"#. Asks the scheduler, with the existing"},{"line_number":213,"context_line":"   ``SchedulerAPI.validate_host_capacity()`` RPC (also used when a"},{"line_number":214,"context_line":"   group is created from a source), whether, in the scheduler\u0027s state"},{"line_number":215,"context_line":"   for this pool (``cluster@backend#pool`` in a cluster), a volume of"}],"source_content_type":"text/x-rst","patch_set":7,"id":"c3b8d299_f3fdf3e4","line":212,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: the Handler names the context for every other step that leaves the process (the image lookup and the create carry the `[glance]` token, the delete and the `cinder-manage` RPC do not) but not this one. `serialize_context` sends `context.to_dict()`, which carries `auth_token` and `service_catalog`, so the choice decides whether the token also travels to the scheduler, which elevates the context itself and needs none. Say \"with the token-less context received from `cinder-manage`\" and mirror it after 322.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":209,"context_line":"   image.min_disk, 1)``, so that the create flow accepts it"},{"line_number":210,"context_line":"   (``image_utils.check_virtual_size()``) and it meets the image\u0027s"},{"line_number":211,"context_line":"   ``min_disk``."},{"line_number":212,"context_line":"#. Asks the scheduler, with the existing"},{"line_number":213,"context_line":"   ``SchedulerAPI.validate_host_capacity()`` RPC (also used when a"},{"line_number":214,"context_line":"   group is created from a source), whether, in the scheduler\u0027s state"},{"line_number":215,"context_line":"   for this pool (``cluster@backend#pool`` in a cluster), a volume of"}],"source_content_type":"text/x-rst","patch_set":7,"id":"c05a8ada_ff88ca31","line":212,"in_reply_to":"c3b8d299_f3fdf3e4","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":243,"context_line":"   gone, since ``delete_volume()`` skips a busy volume without"},{"line_number":244,"context_line":"   raising. The deletion releases its quota; the cache volume owned"},{"line_number":245,"context_line":"   by the internal tenant remains."},{"line_number":246,"context_line":"#. Looks the entry up again the same way. A successful create does"},{"line_number":247,"context_line":"   not guarantee an entry, so a miss returns ``failed`` with \"volume"},{"line_number":248,"context_line":"   was created but no cache entry was recorded\". No entry is recorded"},{"line_number":249,"context_line":"   when the driver clones the image itself (``clone_image()``), when"}],"source_content_type":"text/x-rst","patch_set":7,"id":"69e61de3_7c2c7213","line":246,"updated":"2026-11-01 17:40:12.000000000","message":"The four causes sharing one reason and exit 1 is what I asked for and is right, since the handler cannot tell them apart without a create-flow change. What the spec does not say is that two of them are permanent properties of the backend: where the driver\u0027s `clone_image()` serves the image (RBD with raw images in the same Ceph cluster, NetApp ONTAP NFS, VMware, GPFS with `gpfs_images_share_mode`) or a `cinder://` location is used, `_create_from_image` never reaches the cache, so warm exits 1 on every image there, for good, with the same reason as a broken cache. Please add a sentence under Other deployer impact that `warm` applies only where a create from the image would download it and is not applicable on those backends, and let the reason name the cause set, e.g. \"no cache entry was recorded (driver cloned the image directly, cinder:// location, or the cache refused the entry)\". Keep failed and exit 1; exit 0 would bring back the false success my L116 objected to.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":243,"context_line":"   gone, since ``delete_volume()`` skips a busy volume without"},{"line_number":244,"context_line":"   raising. The deletion releases its quota; the cache volume owned"},{"line_number":245,"context_line":"   by the internal tenant remains."},{"line_number":246,"context_line":"#. Looks the entry up again the same way. A successful create does"},{"line_number":247,"context_line":"   not guarantee an entry, so a miss returns ``failed`` with \"volume"},{"line_number":248,"context_line":"   was created but no cache entry was recorded\". No entry is recorded"},{"line_number":249,"context_line":"   when the driver clones the image itself (``clone_image()``), when"}],"source_content_type":"text/x-rst","patch_set":7,"id":"4a2eb960_5dafe5dc","line":246,"in_reply_to":"69e61de3_7c2c7213","updated":"2026-11-02 12:38:15.000000000","message":"Done. The reason now lists the cause set (316-319: \"possible causes: driver clone_image(), cloning from a cinder:// location, ensure_space() refusal, or cache-volume cloning failure\"), and the result stays failed with exit 1.\n\nThe deployer note (536-542) is worded per image and pool. For example, RBD checks the location\u0027s fsid and requires a raw image before using a direct clone. The note describes successful clone_image() or cinder:// clones, which bypass cache population, rather than treating every image on such a backend as inapplicable. An entry found at the initial lookup still returns exists.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":249,"context_line":"   when the driver clones the image itself (``clone_image()``), when"},{"line_number":250,"context_line":"   the volume is cloned from a ``cinder://`` image location"},{"line_number":251,"context_line":"   (``allowed_direct_url_schemes``), when ``ensure_space()`` cannot"},{"line_number":252,"context_line":"   make room under ``image_volume_cache_max_size_gb`` (e.g. the image"},{"line_number":253,"context_line":"   is larger than the limit), or when cloning the cache volume fails."},{"line_number":254,"context_line":"   Otherwise it returns ``created``."},{"line_number":255,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"d3c8d5a8_88891b34","line":252,"updated":"2026-11-01 17:40:12.000000000","message":"This case is knowable before anything is created. `ensure_space()` runs only in `_create_image_cache_volume_entry` (`volume/manager.py:1576-1601`), reached after the image has been fetched and written, and its first test (`image/cache.py:167-169`) is `volume.size \u003e max_cache_size_gb` with `volume.size` by then set to `ceil(virtual_size)`. Both operands are in the handler\u0027s hands at step 8: the virtual size from Glance and `max_cache_size_gb` from the backend\u0027s own `ImageVolumeCache` (`manager.py:349-355`). The spec already refuses encrypted types and scheduler misses before creating anything, so this is the one knowable refusal left to a full download plus a create and delete. Add a step after 208: if `image_volume_cache_max_size_gb` is set and `ceil(image.virtual_size GiB)` exceeds it, fail with that reason without creating anything; skip the check when Glance reports no `virtual_size`, and leave `max_count` alone since `ensure_space` evicts for it.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":249,"context_line":"   when the driver clones the image itself (``clone_image()``), when"},{"line_number":250,"context_line":"   the volume is cloned from a ``cinder://`` image location"},{"line_number":251,"context_line":"   (``allowed_direct_url_schemes``), when ``ensure_space()`` cannot"},{"line_number":252,"context_line":"   make room under ``image_volume_cache_max_size_gb`` (e.g. the image"},{"line_number":253,"context_line":"   is larger than the limit), or when cloning the cache volume fails."},{"line_number":254,"context_line":"   Otherwise it returns ``created``."},{"line_number":255,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"73d4f759_9b39b111","line":252,"in_reply_to":"d3c8d5a8_88891b34","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":253,"context_line":"   is larger than the limit), or when cloning the cache volume fails."},{"line_number":254,"context_line":"   Otherwise it returns ``created``."},{"line_number":255,"context_line":""},{"line_number":256,"context_line":"If the transient volume is still there after the delete, the result"},{"line_number":257,"context_line":"is ``failed`` (exit 1) naming it, and the cache entry if one was"},{"line_number":258,"context_line":"recorded: the pool may be warm, but the volume and its quota need"},{"line_number":259,"context_line":"cleaning up."}],"source_content_type":"text/x-rst","patch_set":7,"id":"c574bd12_d7c0fa58","line":256,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: 241 and 256-259 cover a transient that survives the in-run delete, not cinder-volume stopping mid-create. The transient then gets the standard restart handling of any volume in `creating`: `_do_cleanup` (`volume/manager.py:718-722`) sets it to `error` and does not delete it, its committed quota stays held, and the CLI sees only the `--timeout` failure. Inherited behaviour, so one sentence after 259 is enough: on a service stop during the create the transient is set to `error` by start-up cleanup, not deleted, and the operator removes `cinder-image-cache-warm-\u003cimage-id\u003e` from the internal tenant; a cache entry may or may not have been recorded.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":253,"context_line":"   is larger than the limit), or when cloning the cache volume fails."},{"line_number":254,"context_line":"   Otherwise it returns ``created``."},{"line_number":255,"context_line":""},{"line_number":256,"context_line":"If the transient volume is still there after the delete, the result"},{"line_number":257,"context_line":"is ``failed`` (exit 1) naming it, and the cache entry if one was"},{"line_number":258,"context_line":"recorded: the pool may be warm, but the volume and its quota need"},{"line_number":259,"context_line":"cleaning up."}],"source_content_type":"text/x-rst","patch_set":7,"id":"3e3343e7_c7824369","line":256,"in_reply_to":"c574bd12_d7c0fa58","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":258,"context_line":"recorded: the pool may be warm, but the volume and its quota need"},{"line_number":259,"context_line":"cleaning up."},{"line_number":260,"context_line":""},{"line_number":261,"context_line":"Concurrent ``warm`` runs for the same image and pool do not create"},{"line_number":262,"context_line":"duplicate entries: the create flow holds a coordination lock on the"},{"line_number":263,"context_line":"image id around the cache populate and looks the entry up again inside"},{"line_number":264,"context_line":"it, so the image is downloaded once and one entry is recorded. Each"}],"source_content_type":"text/x-rst","patch_set":7,"id":"f9f524d0_6d3a01d6","line":261,"updated":"2026-11-01 17:40:12.000000000","message":"Still open from PS6 (my L123). This paragraph covers the same image and pool, but the lock is `@coordination.synchronized(\u0027{image_id}\u0027)` on `_prepare_image_cache_entry` (`flows/manager/create_volume.py:874`), keyed on the image alone, with the Glance download and the cache clone inside it, and the in-lock `get_entry` is filtered by host or `cluster_name` (`image/cache.py:58-63`). With the default `file://$state_path` backend (`coordination.py:40`) every backend process of a host shares the lock, and in a cluster every member does, so the per-pool runs 106-107 prescribes serialise: the k-th run waits roughly k downloads, and with the 600 second default at 143 the later ones exit 1 while the backend keeps working. Please say so, note that a driver reporting `clone_across_pools` is the exception (later runs find the entry inside the lock and return `exists`), and tell operators to run the per-pool warms of one image in sequence or raise `--timeout`; 378-382 should then read \"for each concurrent warm of a different image\".","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":258,"context_line":"recorded: the pool may be warm, but the volume and its quota need"},{"line_number":259,"context_line":"cleaning up."},{"line_number":260,"context_line":""},{"line_number":261,"context_line":"Concurrent ``warm`` runs for the same image and pool do not create"},{"line_number":262,"context_line":"duplicate entries: the create flow holds a coordination lock on the"},{"line_number":263,"context_line":"image id around the cache populate and looks the entry up again inside"},{"line_number":264,"context_line":"it, so the image is downloaded once and one entry is recorded. Each"}],"source_content_type":"text/x-rst","patch_set":7,"id":"d8d97bf0_cd25ecaf","line":261,"in_reply_to":"f9f524d0_6d3a01d6","updated":"2026-11-02 12:38:15.000000000","message":"Added at 360-371: the lock is keyed on the image alone and covers the download and the cache clone, so runs of one image on several pools wait for each other, and operators are told to run them in sequence or raise --timeout. The lock scope is stated by what actually shares it: with file://$state_path, the processes that share its lock directory; with a distributed backend, the processes using that backend.\n\nTwo points differ from the suggestion:\n\n- clone_across_pools: a later run whose first lookup missed reports\n  \"created\", not \"exists\". It has already created its transient volume\n  when it waits on the lock, then finds the first entry inside the lock\n  and clones from it (355-356, 368-371).\n\n- Quota (487-494): quota is reserved and the transient volume is created\n  before create_volume() reaches the image lock. Runs waiting on that\n  lock therefore still hold their transient volumes, even for the same\n  image. I kept the general rule: one transient volume per run that\n  creates one, plus one cache volume per recorded entry.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":330,"context_line":"  (``ExtractVolumeRequestTask._get_image_metadata()``) before it is"},{"line_number":331,"context_line":"  scheduled, so a user who cannot see the image fails before the"},{"line_number":332,"context_line":"  cache is involved."},{"line_number":333,"context_line":"* ``list`` resolves image names with a token it obtains from"},{"line_number":334,"context_line":"  ``[keystone_authtoken]`` inside the ``cinder-manage`` process; it"},{"line_number":335,"context_line":"  makes no RPC call."},{"line_number":336,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"faad16da_f68f3430","line":333,"updated":"2026-11-01 17:40:12.000000000","message":"Lines 315-317 justify `[glance]` for `warm` as the section Cinder already uses for Glance, yet `list`, which makes the same `image.show` calls, is given a second credential set here with no reason. `[keystone_authtoken]` as an outgoing identity has precedent (`api/api_utils.py:183`, `keymgr/migration.py:91`), so this is coherence rather than correctness, but one category needing two sets of service credentials wants a sentence. The larger gap is the endpoint: `list` runs on `get_admin_context()`, whose `service_catalog` is empty (`context.py:135`), and `get_api_servers()` reads the catalog when `glance_api_servers` is unset (`image/glance.py:164-190`), so without `[DEFAULT] glance_api_servers` on the host running `cinder-manage` every name prints `\u003cnot found\u003e` with nothing to say why; Other deployer impact states that requirement for the target cinder-volume only. Simplest: have `list` load its auth from `[glance]` too, reword 333-335, and add a bullet for the `cinder-manage` host: `[glance]` credentials and `glance_api_servers`, or the names show `\u003cnot found\u003e`.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":330,"context_line":"  (``ExtractVolumeRequestTask._get_image_metadata()``) before it is"},{"line_number":331,"context_line":"  scheduled, so a user who cannot see the image fails before the"},{"line_number":332,"context_line":"  cache is involved."},{"line_number":333,"context_line":"* ``list`` resolves image names with a token it obtains from"},{"line_number":334,"context_line":"  ``[keystone_authtoken]`` inside the ``cinder-manage`` process; it"},{"line_number":335,"context_line":"  makes no RPC call."},{"line_number":336,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"e4ae8603_d4ed1ea2","line":333,"in_reply_to":"faad16da_f68f3430","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":354,"context_line":""},{"line_number":355,"context_line":"No new notifications. The transient volume sends the existing"},{"line_number":356,"context_line":"``volume.create.*`` / ``volume.delete.*``. The lookups before and"},{"line_number":357,"context_line":"after the create send the existing ``image_volume_cache.miss`` /"},{"line_number":358,"context_line":"``image_volume_cache.hit``, as the create\u0027s own lookups do, so cache"},{"line_number":359,"context_line":"hit rates derived from them include ``warm`` runs."},{"line_number":360,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"5d4aef05_ad64118e","line":357,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: `get_entry()` goes through `image_volume_cache_get_and_update_last_used` (`db/api.py:8496`), so every `warm`, including an `exists` run, moves the entry to the front of the LRU and defers its eviction; 415-417 should say so. And the counts are asymmetric: a user create miss emits three misses, a warm miss emits four plus one hit from the post-create check, so hit rates derived from these notifications should discount `warm` runs rather than \"include\" them.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":354,"context_line":""},{"line_number":355,"context_line":"No new notifications. The transient volume sends the existing"},{"line_number":356,"context_line":"``volume.create.*`` / ``volume.delete.*``. The lookups before and"},{"line_number":357,"context_line":"after the create send the existing ``image_volume_cache.miss`` /"},{"line_number":358,"context_line":"``image_volume_cache.hit``, as the create\u0027s own lookups do, so cache"},{"line_number":359,"context_line":"hit rates derived from them include ``warm`` runs."},{"line_number":360,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"85284f2f_8845d994","line":357,"in_reply_to":"5d4aef05_ad64118e","updated":"2026-11-02 12:38:15.000000000","message":"Done at 549-551 for last_used. For the counts (462-468) I wrote that they include warm\u0027s extra lookups and must not be read directly as the cache efficiency of user creates, rather than asking readers to discount warm runs: nothing in the notifications marks a warm run, and the number of lookups a run makes depends on its path (exists, created, or waiting on the lock), so there is no fixed amount to take off.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"22c77070ff3fcb78203fb2453390c177cc07ce54","unresolved":true,"context_lines":[{"line_number":405,"context_line":"  * A Glance endpoint in ``[DEFAULT] glance_api_servers``. The"},{"line_number":406,"context_line":"    internal-tenant context carries no service catalog, so without it"},{"line_number":407,"context_line":"    the image lookup fails."},{"line_number":408,"context_line":"  * ``[DEFAULT] default_volume_type`` when ``--volume-type`` is not"},{"line_number":409,"context_line":"    given."},{"line_number":410,"context_line":""},{"line_number":411,"context_line":"* Takes effect after upgrade; nothing to enable beyond the"}],"source_content_type":"text/x-rst","patch_set":7,"id":"8dd3f025_258c8cc4","line":408,"updated":"2026-11-01 17:40:12.000000000","message":"Nit: `default_volume_type` is `required\u003dTrue` with default `__DEFAULT__` (`cinder/common/config.py:99-102`), so it always has a value and this bullet names nothing a deployer has to configure. The real requirement is the one the Handler states at 204-207 and 212-221: the type must exist, must not be encrypted and must pass the pool\u0027s filters, since `__DEFAULT__` can carry extra specs. Reword to that.","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"5f048b54276b46a58e9ebf8d490e110b11fea5b7","unresolved":false,"context_lines":[{"line_number":405,"context_line":"  * A Glance endpoint in ``[DEFAULT] glance_api_servers``. The"},{"line_number":406,"context_line":"    internal-tenant context carries no service catalog, so without it"},{"line_number":407,"context_line":"    the image lookup fails."},{"line_number":408,"context_line":"  * ``[DEFAULT] default_volume_type`` when ``--volume-type`` is not"},{"line_number":409,"context_line":"    given."},{"line_number":410,"context_line":""},{"line_number":411,"context_line":"* Takes effect after upgrade; nothing to enable beyond the"}],"source_content_type":"text/x-rst","patch_set":7,"id":"2c3dc57f_e4af5526","line":408,"in_reply_to":"8dd3f025_258c8cc4","updated":"2026-11-02 12:38:15.000000000","message":"Done","commit_id":"733e94dc367fe9cfb682d011edb8ccab0fbe63f1"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":95,"context_line":"``cluster@backend#pool`` for a clustered (Active/Active) backend."},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"* A ``--host`` without a pool part exits 2."},{"line_number":98,"context_line":"* A ``--host`` whose part before ``#`` is not a ``cinder-volume``"},{"line_number":99,"context_line":"  service or cluster exits 2."},{"line_number":100,"context_line":"* Specifying an individual cluster member causes the command to exit"},{"line_number":101,"context_line":"  with code 2, and the message names the cluster\u0027s pool to use"}],"source_content_type":"text/x-rst","patch_set":8,"id":"80b395c8_fcf9b2d6","line":98,"updated":"2026-11-10 19:36:25.000000000","message":"These two bullets let `node1#pool` through: the part before `#` is `node1`, and `db.service_get(host\u003d\u0027node1\u0027)` picks match level host at `db/api.py:697-723`, which matches `node1@lvm` with a LIKE, so the value is accepted and the service is up. The call is then addressed to topic `cinder-volume.node1`, which nothing serves, and the run waits the full `--timeout` before exiting 1. 986740 already rejects a `--host` without `@`; please state it: a `--host` without a backend part (`@`) or a pool part (`#`) exits 2.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":95,"context_line":"``cluster@backend#pool`` for a clustered (Active/Active) backend."},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"* A ``--host`` without a pool part exits 2."},{"line_number":98,"context_line":"* A ``--host`` whose part before ``#`` is not a ``cinder-volume``"},{"line_number":99,"context_line":"  service or cluster exits 2."},{"line_number":100,"context_line":"* Specifying an individual cluster member causes the command to exit"},{"line_number":101,"context_line":"  with code 2, and the message names the cluster\u0027s pool to use"}],"source_content_type":"text/x-rst","patch_set":8,"id":"1d7d35a7_637bb813","line":98,"in_reply_to":"80b395c8_fcf9b2d6","updated":"2026-11-11 06:32:35.000000000","message":"986740 now also checks that the \u0027@\u0027 is in the backend part, so host#pool@fast exits 2 as well.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":100,"context_line":"* Specifying an individual cluster member causes the command to exit"},{"line_number":101,"context_line":"  with code 2, and the message names the cluster\u0027s pool to use"},{"line_number":102,"context_line":"  instead."},{"line_number":103,"context_line":"* If nothing is up for the service or cluster, the pool is reported"},{"line_number":104,"context_line":"  as failed (exit 1) without sending anything."},{"line_number":105,"context_line":"* A backend with several pools is warmed by running ``warm`` once per"},{"line_number":106,"context_line":"  pool, except a non-clustered backend whose driver reports"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7c16fd8d_fab578ce","line":103,"updated":"2026-11-10 19:36:25.000000000","message":"Nit: `Service.is_up` and `Cluster.is_up` look only at the heartbeat, `objects/service.py:211-214` and `objects/cluster.py:161-163`, so a disabled or frozen service passes this bullet and the call is sent. The scheduler builds its state from services with `disabled\u003dFalse` and `frozen\u003dFalse` at `scheduler/host_manager.py:686-689`, so step 9 at 269-279 then fails with the placement reason, which the text frames as a type question. Either add \"disabled or frozen\" here or say at 279 that such a backend is absent from the scheduler\u0027s state and fails there.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":100,"context_line":"* Specifying an individual cluster member causes the command to exit"},{"line_number":101,"context_line":"  with code 2, and the message names the cluster\u0027s pool to use"},{"line_number":102,"context_line":"  instead."},{"line_number":103,"context_line":"* If nothing is up for the service or cluster, the pool is reported"},{"line_number":104,"context_line":"  as failed (exit 1) without sending anything."},{"line_number":105,"context_line":"* A backend with several pools is warmed by running ``warm`` once per"},{"line_number":106,"context_line":"  pool, except a non-clustered backend whose driver reports"}],"source_content_type":"text/x-rst","patch_set":8,"id":"55bf747c_4b1635c0","line":103,"in_reply_to":"7c16fd8d_fab578ce","updated":"2026-11-11 06:32:35.000000000","message":"Done. Added that the check looks only at the heartbeat, and that a disabled or frozen service or cluster then fails the scheduler check on a cache miss.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":162,"context_line":"  ``clone_across_pools`` it can be another pool. A ``failed`` result"},{"line_number":163,"context_line":"  without an entry carries the requested pool."},{"line_number":164,"context_line":""},{"line_number":165,"context_line":"``warm`` disables ``[oslo_messaging_rabbit] use_queue_manager`` for"},{"line_number":166,"context_line":"its own process. With it, a reply queue is named"},{"line_number":167,"context_line":"``\u003chostname\u003e:\u003cprocessname\u003e:\u003ccounter\u003e`` and the counter restarts for"},{"line_number":168,"context_line":"each new process group, so two ``cinder-manage`` runs at the same time"}],"source_content_type":"text/x-rst","patch_set":8,"id":"43cbf057_275df872","line":165,"updated":"2026-11-10 19:36:25.000000000","message":"Nit: say this happens before the RPC transport is initialised. The rabbit driver reads the option when the transport is created, `impl_rabbit.py:1733-1738` in oslo.messaging 18 and again per connection at 887-892, which cinder does in `rpc.init` at `rpc.py:58-62`. The existing volume delete calls `rpc.init(CONF)` right before its RPC at `cmd/manage.py:619`, so an override placed after it, the natural spot when copying that pattern, is a silent no-op. 986740 orders it correctly; one clause here pins it.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":162,"context_line":"  ``clone_across_pools`` it can be another pool. A ``failed`` result"},{"line_number":163,"context_line":"  without an entry carries the requested pool."},{"line_number":164,"context_line":""},{"line_number":165,"context_line":"``warm`` disables ``[oslo_messaging_rabbit] use_queue_manager`` for"},{"line_number":166,"context_line":"its own process. With it, a reply queue is named"},{"line_number":167,"context_line":"``\u003chostname\u003e:\u003cprocessname\u003e:\u003ccounter\u003e`` and the counter restarts for"},{"line_number":168,"context_line":"each new process group, so two ``cinder-manage`` runs at the same time"}],"source_content_type":"text/x-rst","patch_set":8,"id":"a1f0b4f1_ee495bbf","line":165,"in_reply_to":"43cbf057_275df872","updated":"2026-11-11 06:32:35.000000000","message":"Done","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":226,"context_line":"   (``driver.get_volume_stats(refresh\u003dFalse)``), or, if it reports no"},{"line_number":227,"context_line":"   ``pools`` list, its ``volume_backend_name``, or ``_pool0`` if that"},{"line_number":228,"context_line":"   is unset. Otherwise it fails and lists them."},{"line_number":229,"context_line":"#. Looks the entry up with ``ImageVolumeCache.get_entry()`` the way"},{"line_number":230,"context_line":"   the create flow does for a volume on the pool, so a hit is the"},{"line_number":231,"context_line":"   entry a create on the pool would try to clone. In a cluster the"},{"line_number":232,"context_line":"   lookup is limited to ``cluster@backend#pool``; otherwise to"}],"source_content_type":"text/x-rst","patch_set":8,"id":"251851ee_77609bb3","line":229,"updated":"2026-11-10 19:36:25.000000000","message":"Nit: every other call that leaves the process names its context, at 216-219, 271-272 and 304-305, but the two `get_entry()` lookups do not. \"The way the create flow does\" points at the internal-tenant context, `flows/manager/create_volume.py:884-891`, which at step 6 already carries the `[glance]` token, while 986740 passes the token-less context from `cinder-manage` to both. The choice is visible: `get_entry()` sends the hit and miss notifications with it at `image/cache.py:94-98` and evicts an outdated entry through `volume_api.delete(context, volume)` at 88-90 and 147-150. Say \"both lookups use the token-less context received from `cinder-manage`, and an outdated entry a lookup finds is evicted with it\", and mirror it after 428-429.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":226,"context_line":"   (``driver.get_volume_stats(refresh\u003dFalse)``), or, if it reports no"},{"line_number":227,"context_line":"   ``pools`` list, its ``volume_backend_name``, or ``_pool0`` if that"},{"line_number":228,"context_line":"   is unset. Otherwise it fails and lists them."},{"line_number":229,"context_line":"#. Looks the entry up with ``ImageVolumeCache.get_entry()`` the way"},{"line_number":230,"context_line":"   the create flow does for a volume on the pool, so a hit is the"},{"line_number":231,"context_line":"   entry a create on the pool would try to clone. In a cluster the"},{"line_number":232,"context_line":"   lookup is limited to ``cluster@backend#pool``; otherwise to"}],"source_content_type":"text/x-rst","patch_set":8,"id":"5f1a5545_49c061ac","line":229,"in_reply_to":"251851ee_77609bb3","updated":"2026-11-11 06:32:35.000000000","message":"Done, in step 6 and in Security impact.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":236,"context_line":"   not check that its cache volume exists or is usable. If the cache"},{"line_number":237,"context_line":"   volume still has its volume record but is unusable, delete it"},{"line_number":238,"context_line":"   through the volume API, which evicts the entry, and run ``warm``"},{"line_number":239,"context_line":"   again. Deleting it directly in the database or with"},{"line_number":240,"context_line":"   ``cinder-manage volume delete`` skips the eviction and leaves the"},{"line_number":241,"context_line":"   entry in place."},{"line_number":242,"context_line":"#. Picks the volume type: ``volume_type_id`` if given, else"}],"source_content_type":"text/x-rst","patch_set":8,"id":"e8a94266_d6b22342","line":239,"updated":"2026-11-10 19:36:25.000000000","message":"True, but this names a dead end without a way out, since the remedy at 237-239 needs the volume record. Once the row is gone nothing in cinder removes the entry: `volume_destroy` touches only `VOLUME_DEPENDENT_MODELS` at `db/api.py:2106-2115`, `image_volume_cache_delete` is reached only through `evict()` at `image/cache.py:54`, and a create does not heal it: the clone from the entry fails with `VolumeNotFound` at `flows/manager/create_volume.py:656`, is swallowed at 970 into a download, and the populate then returns at `manager.py:1588-1592` because `get_entry` still finds the row. So `warm` keeps saying exists while every create downloads, and once the image is updated the eviction in `delete_cached_volume` raises `VolumeNotFound` from `objects.Volume.get_by_id` at `image/cache.py:147` inside `_prepare_image_cache_entry`, and every create from that image on the pool fails. Please say so here, with the only remedy there is today, deleting the row from `image_volume_cache_entries`, or reconsider the `evict` action excluded at 63-65 for exactly this case.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":236,"context_line":"   not check that its cache volume exists or is usable. If the cache"},{"line_number":237,"context_line":"   volume still has its volume record but is unusable, delete it"},{"line_number":238,"context_line":"   through the volume API, which evicts the entry, and run ``warm``"},{"line_number":239,"context_line":"   again. Deleting it directly in the database or with"},{"line_number":240,"context_line":"   ``cinder-manage volume delete`` skips the eviction and leaves the"},{"line_number":241,"context_line":"   entry in place."},{"line_number":242,"context_line":"#. Picks the volume type: ``volume_type_id`` if given, else"}],"source_content_type":"text/x-rst","patch_set":8,"id":"a666c5cc_e4f22e9b","line":239,"in_reply_to":"e8a94266_d6b22342","updated":"2026-11-11 06:32:35.000000000","message":"Done. Documented the dead end and the only remedy, deleting the entry\u0027s row, rather than reintroducing evict. Scoped to creates whose cache lookup finds the entry, since with clone_across_pools that can be another pool.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":313,"context_line":"   metadata read at the start, an image updated in between would make"},{"line_number":314,"context_line":"   this lookup treat the new entry as outdated and delete it. Fails if"},{"line_number":315,"context_line":"   getting the token or reading the metadata fails. A successful create"},{"line_number":316,"context_line":"   does not guarantee an entry, so a miss returns ``failed`` with"},{"line_number":317,"context_line":"   \"volume was created but no cache entry was recorded (possible"},{"line_number":318,"context_line":"   causes: driver clone_image(), cloning from a cinder:// location,"},{"line_number":319,"context_line":"   ensure_space() refusal, or cache-volume cloning failure)\". No entry"}],"source_content_type":"text/x-rst","patch_set":8,"id":"e91d0745_f93be8d5","line":316,"updated":"2026-11-10 19:36:25.000000000","message":"One cause is missing, and it is the one the re-read at 309-311 itself introduces. `_should_update_entry` is an inequality test at `image/cache.py:273`, so an image updated between the create flow\u0027s own metadata read at `flows/manager/create_volume.py:325`, the value the entry stores, and this re-read makes the final `get_entry()` delete the entry just recorded, cache volume included, and this reason is returned with none of its four causes true. 311-314 cover only an update before the flow\u0027s read. Add a fifth cause here and at 320-326, \"or the image was updated during the run, so the final lookup found the new entry outdated and deleted it; run `warm` again\", and update the string in 986740.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":313,"context_line":"   metadata read at the start, an image updated in between would make"},{"line_number":314,"context_line":"   this lookup treat the new entry as outdated and delete it. Fails if"},{"line_number":315,"context_line":"   getting the token or reading the metadata fails. A successful create"},{"line_number":316,"context_line":"   does not guarantee an entry, so a miss returns ``failed`` with"},{"line_number":317,"context_line":"   \"volume was created but no cache entry was recorded (possible"},{"line_number":318,"context_line":"   causes: driver clone_image(), cloning from a cinder:// location,"},{"line_number":319,"context_line":"   ensure_space() refusal, or cache-volume cloning failure)\". No entry"}],"source_content_type":"text/x-rst","patch_set":8,"id":"9ef8caa4_5035df25","line":316,"in_reply_to":"e91d0745_f93be8d5","updated":"2026-11-11 06:32:35.000000000","message":"Done. Added to the cause set and the string, and as a separate sentence after the list, since here the entry is recorded and then deleted. String updated in 986740.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":432,"context_line":"  Kolla gives the cinder service user, it can read private images of"},{"line_number":433,"context_line":"  any project, so ``warm`` is only exposed through ``cinder-manage``,"},{"line_number":434,"context_line":"  which requires access to the cinder configuration and database."},{"line_number":435,"context_line":"* A cache entry does not widen access to an image. A user create"},{"line_number":436,"context_line":"  still looks the image up with the user\u0027s own token in the API"},{"line_number":437,"context_line":"  (``ExtractVolumeRequestTask._get_image_metadata()``) before it is"},{"line_number":438,"context_line":"  scheduled, so a user who cannot see the image fails before the"}],"source_content_type":"text/x-rst","patch_set":8,"id":"501185a4_135b61eb","line":435,"updated":"2026-11-10 19:36:25.000000000","message":"This holds for visibility, not for Glance\u0027s `download_image` policy. A user create checks the image with the user\u0027s token in the API at `flows/api/create_volume.py:249-251`, and on a cache hit the manager clones the cache volume without any Glance call, so `download_image` is never evaluated for that user. The existing cache already allows that once a permitted user has created from the image; `warm` creates the entry with the `[glance]` identity instead, so a user who can see the image but is denied its download gets the data on their first create. Qualify the bullet to \"does not widen visibility\" and add: do not warm images whose download is restricted by Glance policy.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":432,"context_line":"  Kolla gives the cinder service user, it can read private images of"},{"line_number":433,"context_line":"  any project, so ``warm`` is only exposed through ``cinder-manage``,"},{"line_number":434,"context_line":"  which requires access to the cinder configuration and database."},{"line_number":435,"context_line":"* A cache entry does not widen access to an image. A user create"},{"line_number":436,"context_line":"  still looks the image up with the user\u0027s own token in the API"},{"line_number":437,"context_line":"  (``ExtractVolumeRequestTask._get_image_metadata()``) before it is"},{"line_number":438,"context_line":"  scheduled, so a user who cannot see the image fails before the"}],"source_content_type":"text/x-rst","patch_set":8,"id":"8ebdc6ed_b6464d2c","line":435,"in_reply_to":"501185a4_135b61eb","updated":"2026-11-11 06:32:35.000000000","message":"Done","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":441,"context_line":"  with a token from the same ``[glance]`` service credentials; it"},{"line_number":442,"context_line":"  makes no RPC call."},{"line_number":443,"context_line":""},{"line_number":444,"context_line":"The handler reuses the existing image download and conversion path;"},{"line_number":445,"context_line":"no new sudo use or cryptography is introduced."},{"line_number":446,"context_line":""},{"line_number":447,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"0b771f3d_1742c90e","line":444,"updated":"2026-11-10 19:36:25.000000000","message":"Literally true, but the reused path verifies a signed image\u0027s signature as a different identity, and nothing under Other deployer impact says so. With `verify_glance_signatures` enabled, `flows/manager/create_volume.py:994-1001` verifies with the flow context, `image_utils.py:673` hands it to cursive, which fetches the signing certificate through the key manager with `context.get_auth_plugin()` from `context.py:147-151`, and castellan resolves the Barbican endpoint from the catalog unless `[barbican] barbican_endpoint` is set. In a warm that context is the internal tenant carrying the `[glance]` token and no catalog, so the certificate owned by the signer\u0027s project is read as the `[glance]` identity, which Barbican\u0027s project-scoped policy refuses without an ACL, and the endpoint lookup fails; both become `ImageSignatureVerificationException` at 1033-1040 and `warm` reports \"transient volume creation failed\" for an image its owner creates from successfully. Please add a deployer bullet: a signed image is verified with the `[glance]` identity, which must be able to read the image\u0027s signing certificate in the key manager, and `[barbican] barbican_endpoint` must be set.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":441,"context_line":"  with a token from the same ``[glance]`` service credentials; it"},{"line_number":442,"context_line":"  makes no RPC call."},{"line_number":443,"context_line":""},{"line_number":444,"context_line":"The handler reuses the existing image download and conversion path;"},{"line_number":445,"context_line":"no new sudo use or cryptography is introduced."},{"line_number":446,"context_line":""},{"line_number":447,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"6e24e716_28a7607a","line":444,"in_reply_to":"0b771f3d_1742c90e","updated":"2026-11-11 06:32:35.000000000","message":"Done. Added under the warm requirements, limited to runs that download a signed image, and to the Barbican key manager for barbican_endpoint.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"9234929351e4f229e9e564d11834e68801b9d9fd","unresolved":true,"context_lines":[{"line_number":490,"context_line":"  entry recorded adds a cache volume of the image\u0027s virtual size,"},{"line_number":491,"context_line":"  rounded up to GiB, which remains afterwards. Concurrent ``warm``"},{"line_number":492,"context_line":"  runs therefore need quota, including the per-type quotas, for one"},{"line_number":493,"context_line":"  transient volume per run plus one cache volume per entry they"},{"line_number":494,"context_line":"  record."},{"line_number":495,"context_line":"* On acceptance, ``validate_host_capacity()`` accounts for the"},{"line_number":496,"context_line":"  requested size in the scheduler\u0027s in-memory pool state. If ``warm``"}],"source_content_type":"text/x-rst","patch_set":8,"id":"57fc59ca_fb8aca4c","line":493,"updated":"2026-11-10 19:36:25.000000000","message":"Say what happens when the internal tenant has room for the transient but not for the cache volume, because the two are checked at different points. The transient\u0027s reservation at 280-282 comes before anything is created; the cache volume\u0027s is the `QUOTAS.reserve` at `volume/manager.py:1632`, outside the try in `_clone_image_volume` and reached only after the full download, and the OverQuota it raises is a `CinderException`, `exception.py:559`, caught at `manager.py:1618-1620` with a warning. The create succeeds, no entry is recorded and `warm` returns the 316-319 reason, whose cause set does not name quota. With the default `quota_volumes` of 10 this starts at nine cache volumes. One sentence here, and \"internal-tenant quota for the cache volume\" added to the cause set at 316-319 and to the string in 986740.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"},{"author":{"_account_id":34539,"name":"minwoo seo","email":"alsdn1211@naver.com","username":"minwooseo"},"change_message_id":"74ed225fcd81a90cb78c17501315bf68ab42c9f7","unresolved":false,"context_lines":[{"line_number":490,"context_line":"  entry recorded adds a cache volume of the image\u0027s virtual size,"},{"line_number":491,"context_line":"  rounded up to GiB, which remains afterwards. Concurrent ``warm``"},{"line_number":492,"context_line":"  runs therefore need quota, including the per-type quotas, for one"},{"line_number":493,"context_line":"  transient volume per run plus one cache volume per entry they"},{"line_number":494,"context_line":"  record."},{"line_number":495,"context_line":"* On acceptance, ``validate_host_capacity()`` accounts for the"},{"line_number":496,"context_line":"  requested size in the scheduler\u0027s in-memory pool state. If ``warm``"}],"source_content_type":"text/x-rst","patch_set":8,"id":"ce040a2c_50abb30d","line":493,"in_reply_to":"57fc59ca_fb8aca4c","updated":"2026-11-11 06:32:35.000000000","message":"Done. Cause set and string updated, string also in 986740.","commit_id":"393a8c323fea5d1bc0959dae21b9736aaf25e034"}]}
