)]}'
{"/PATCHSET_LEVEL":[{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"357465d501e04f1aac24c2ea7cbc52e16a434c1f","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":7,"id":"efe6b151_84c6ee61","updated":"2026-10-15 12:32:28.000000000","message":"recheck","commit_id":"532483097d7856c4bc1b62142d9454cfe5908850"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":8,"id":"9c0bc18f_ffe204cc","updated":"2026-10-15 15:04:41.000000000","message":"Thanks, this addresses a real gap and the shape is right: ranges on the type, values on the volume, scheduler gating through an unscoped extra spec, driver-first-then-DB on update. It is also clearly written.\n\nMy comments are about semantics rather than structure. Three of them (the range-maximum fallback, what \"equivalent keys\" means for vendor QoS specs, and clearing values on retype) are user-visible behaviour that would be hard to change after implementation, so I\u0027d like them settled in the spec. The rest are clarifications. Speaking as a driver maintainer, implementing `update_volume_qos` on FlashArray would be small, so I\u0027m supportive of the direction.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":5314,"name":"Brian Rosmaita","email":"rosmaita.fossdev@gmail.com","username":"brian-rosmaita"},"change_message_id":"d75f7aa99ae61f023fccd89ad2c50c507d9f430a","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":8,"id":"9580bdcc_77b404aa","updated":"2026-10-16 13:34:57.000000000","message":"We discussed this at the Indri PTG; conclusion was that\n(1) need more clarity on the use cases to be addressed\n(2) would be worth exploring decoupling cinder\u0027s current QoS specs from volume types so that QoS on a type could be changed without having to retype all the volumes; manila did something like this recently, so please look it over and see if a similar approach will meet your use cases: https://specs.openstack.org/openstack//manila-specs/specs/gazpacho/qos_types.html","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"355316b81e3d54e5562c69709a04578de4bf4586","unresolved":true,"context_lines":[],"source_content_type":"","patch_set":8,"id":"4d4d013a_44cf410c","in_reply_to":"2af56e0e_4782dbdf","updated":"2026-11-17 09:01:39.000000000","message":"PS11 also covers the Cinder extension of the Manila model in Alternatives.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":true,"context_lines":[],"source_content_type":"","patch_set":8,"id":"2af56e0e_4782dbdf","in_reply_to":"9580bdcc_77b404aa","updated":"2026-10-30 14:34:57.000000000","message":"Manila QoS types still pick QoS only through the share type, and have no API to change QoS of an existing share. So they don\u0027t cover changing QoS per volume. I added two use cases (backup window, billing by QoS) and the Manila approach to Alternatives.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":10,"id":"cd4cdb17_292280d0","updated":"2026-11-01 15:36:21.000000000","message":"One thing I missed on PS8 keeps me at -1: the design assumes administrators can change the `per_volume_qos:*` extra specs of a type that already has volumes, and Cinder refuses that outright (`cinder/api/contrib/types_extra_specs.py:62-70`). See my comment at line 106; it is the only blocking item, and the rest are clarifications.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"}],"specs/2027.1/per-volume-qos.rst":[{"author":{"_account_id":27615,"name":"Rajat Dhasmana","email":"rajatdhasmana@gmail.com","username":"whoami-rajat"},"change_message_id":"5c1d6792001507fa8ff8fccd304e359b9bcc984d","unresolved":true,"context_lines":[{"line_number":33,"context_line":""},{"line_number":34,"context_line":"* An operator creates a volume with ``max_iops\u003d3000`` and"},{"line_number":35,"context_line":"  ``max_bps\u003d125000000`` without a type for those exact values."},{"line_number":36,"context_line":"* An operator changes the limits of an in-use volume without detaching or"},{"line_number":37,"context_line":"  retyping it, or removes the override to return to the type default."},{"line_number":38,"context_line":"* An operator lets project members tune the limits of their own volumes,"},{"line_number":39,"context_line":"  but only within bounds the operator chooses, for example 3000 to 16000"},{"line_number":40,"context_line":"  IOPS."}],"source_content_type":"text/x-rst","patch_set":8,"id":"bd128198_0bce2694","line":37,"range":{"start_line":36,"start_character":2,"end_line":37,"end_character":13},"updated":"2026-10-15 15:23:43.000000000","message":"I don\u0027t think this is a good idea, an attached volume with ongoing IO should never be subjected to changing the QoS dynamically","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":33,"context_line":""},{"line_number":34,"context_line":"* An operator creates a volume with ``max_iops\u003d3000`` and"},{"line_number":35,"context_line":"  ``max_bps\u003d125000000`` without a type for those exact values."},{"line_number":36,"context_line":"* An operator changes the limits of an in-use volume without detaching or"},{"line_number":37,"context_line":"  retyping it, or removes the override to return to the type default."},{"line_number":38,"context_line":"* An operator lets project members tune the limits of their own volumes,"},{"line_number":39,"context_line":"  but only within bounds the operator chooses, for example 3000 to 16000"},{"line_number":40,"context_line":"  IOPS."}],"source_content_type":"text/x-rst","patch_set":8,"id":"cfca096e_8836e64f","line":37,"range":{"start_line":36,"start_character":2,"end_line":37,"end_character":13},"in_reply_to":"007baaf3_46dc271d","updated":"2026-11-01 15:36:21.000000000","message":"Changing back-end limits on an attached volume is existing Cinder behaviour: NetApp re-provisions the LUN\u0027s policy group on online extend (block_base.py:735-755), Pure re-applies set_qos on in-use extend (pure.py:1764-1783), and in-use retype already changes back-end QoS, since `volume_qos_allows_retype` (api.py:1965) blocks only front-end changes.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":true,"context_lines":[{"line_number":34,"context_line":"* An operator creates a volume with ``max_iops\u003d3000`` and"},{"line_number":35,"context_line":"  ``max_bps\u003d125000000`` without a type for those exact values."},{"line_number":36,"context_line":"* An operator changes the limits of an in-use volume without detaching or"},{"line_number":37,"context_line":"  retyping it, or removes the override to return to the type default."},{"line_number":38,"context_line":"* An operator lets project members tune the limits of their own volumes,"},{"line_number":39,"context_line":"  but only within bounds the operator chooses, for example 3000 to 16000"},{"line_number":40,"context_line":"  IOPS."}],"source_content_type":"text/x-rst","patch_set":8,"id":"007baaf3_46dc271d","line":37,"in_reply_to":"bd128198_0bce2694","updated":"2026-10-30 14:34:57.000000000","message":"Changing the limit only changes the speed of new IO. It does not touch data or the attachment. Cinder already does this: NetApp modifies the QoS policy of an attached LUN on online extend. We also use this with a downstream patch without major issues.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":27615,"name":"Rajat Dhasmana","email":"rajatdhasmana@gmail.com","username":"whoami-rajat"},"change_message_id":"5c1d6792001507fa8ff8fccd304e359b9bcc984d","unresolved":true,"context_lines":[{"line_number":35,"context_line":"  ``max_bps\u003d125000000`` without a type for those exact values."},{"line_number":36,"context_line":"* An operator changes the limits of an in-use volume without detaching or"},{"line_number":37,"context_line":"  retyping it, or removes the override to return to the type default."},{"line_number":38,"context_line":"* An operator lets project members tune the limits of their own volumes,"},{"line_number":39,"context_line":"  but only within bounds the operator chooses, for example 3000 to 16000"},{"line_number":40,"context_line":"  IOPS."},{"line_number":41,"context_line":"* An end user raises the limits of its own volume within the range allowed"},{"line_number":42,"context_line":"  by the volume type, but cannot go beyond that range."}],"source_content_type":"text/x-rst","patch_set":8,"id":"48c313bc_22290c56","line":39,"range":{"start_line":38,"start_character":2,"end_line":39,"end_character":45},"updated":"2026-10-15 15:23:43.000000000","message":"This doesn\u0027t look like a good idea either, exposing the QoS settings to non admin users could lead to users setting the max iops possible causing cotention. Admins make sure there is not a bad actor in the deployment that gets too much control over backend related details.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":true,"context_lines":[{"line_number":36,"context_line":"* An operator changes the limits of an in-use volume without detaching or"},{"line_number":37,"context_line":"  retyping it, or removes the override to return to the type default."},{"line_number":38,"context_line":"* An operator lets project members tune the limits of their own volumes,"},{"line_number":39,"context_line":"  but only within bounds the operator chooses, for example 3000 to 16000"},{"line_number":40,"context_line":"  IOPS."},{"line_number":41,"context_line":"* An end user raises the limits of its own volume within the range allowed"},{"line_number":42,"context_line":"  by the volume type, but cannot go beyond that range."}],"source_content_type":"text/x-rst","patch_set":8,"id":"5355ae65_af8679bd","line":39,"in_reply_to":"48c313bc_22290c56","updated":"2026-10-30 14:34:57.000000000","message":"Users can only choose within a range set by the admin in the volume type, and nobody can go outside it. A deployment that doesn\u0027t want this can make `volume:set_qos` admin only.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"355316b81e3d54e5562c69709a04578de4bf4586","unresolved":true,"context_lines":[{"line_number":36,"context_line":"* An operator changes the limits of an in-use volume without detaching or"},{"line_number":37,"context_line":"  retyping it, or removes the override to return to the type default."},{"line_number":38,"context_line":"* An operator lets project members tune the limits of their own volumes,"},{"line_number":39,"context_line":"  but only within bounds the operator chooses, for example 3000 to 16000"},{"line_number":40,"context_line":"  IOPS."},{"line_number":41,"context_line":"* An end user raises the limits of its own volume within the range allowed"},{"line_number":42,"context_line":"  by the volume type, but cannot go beyond that range."}],"source_content_type":"text/x-rst","patch_set":8,"id":"b60efd13_2c8ff679","line":39,"in_reply_to":"5355ae65_af8679bd","updated":"2026-11-17 09:01:39.000000000","message":"Security impact now covers this (PS11).","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":44,"context_line":"Proposed change"},{"line_number":45,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Add two optional per-volume limits, ``max_iops`` and ``max_bps`` (bytes per"},{"line_number":48,"context_line":"second), stored in new columns of the ``volumes`` table, not in volume"},{"line_number":49,"context_line":"metadata. Both are sustained upper limits for read and write combined. Burst"},{"line_number":50,"context_line":"settings are out of scope: burst limits in the volume type QoS spec still"}],"source_content_type":"text/x-rst","patch_set":8,"id":"9b16d0c8_20b42298","line":47,"updated":"2026-10-15 15:04:41.000000000","message":"Please define `max_bps` precisely: integer bytes per second, no SI or binary prefixes, and drivers round to the backend\u0027s granularity (FlashArray takes bytes per second; some backends take MB/s).","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":44,"context_line":"Proposed change"},{"line_number":45,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Add two optional per-volume limits, ``max_iops`` and ``max_bps`` (bytes per"},{"line_number":48,"context_line":"second), stored in new columns of the ``volumes`` table, not in volume"},{"line_number":49,"context_line":"metadata. Both are sustained upper limits for read and write combined. Burst"},{"line_number":50,"context_line":"settings are out of scope: burst limits in the volume type QoS spec still"}],"source_content_type":"text/x-rst","patch_set":8,"id":"2a5a5d0f_fe9479a0","line":47,"in_reply_to":"9b16d0c8_20b42298","updated":"2026-10-30 14:34:57.000000000","message":"Done.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":66,"context_line":"  ``per_volume_qos:max_iops_min``, ``per_volume_qos:max_iops_max``,"},{"line_number":67,"context_line":"  ``per_volume_qos:max_iops_default``, and the same three keys for"},{"line_number":68,"context_line":"  ``max_bps``. ``_max`` is required for a limit that has any of these keys,"},{"line_number":69,"context_line":"  ``_min`` defaults to 1, and ``_default`` must be between them. The API"},{"line_number":70,"context_line":"  checks this when the extra specs of a type with ``per_volume_qos_support``"},{"line_number":71,"context_line":"  are set. On other types these scoped keys are ignored, as other scoped"},{"line_number":72,"context_line":"  extra specs are."}],"source_content_type":"text/x-rst","patch_set":8,"id":"b38db872_2f181b0c","line":69,"updated":"2026-10-15 15:04:41.000000000","message":"Validating the range keys \"when the extra specs of a type are set\" couples the generic extra-specs API to these keys and is order-dependent (setting `_max` before `_support` skips the check; removing `_max` later leaves a broken type). Please validate at use time (create and `os-set_qos`) as the authoritative check, with the set-time check as a convenience only.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":66,"context_line":"  ``per_volume_qos:max_iops_min``, ``per_volume_qos:max_iops_max``,"},{"line_number":67,"context_line":"  ``per_volume_qos:max_iops_default``, and the same three keys for"},{"line_number":68,"context_line":"  ``max_bps``. ``_max`` is required for a limit that has any of these keys,"},{"line_number":69,"context_line":"  ``_min`` defaults to 1, and ``_default`` must be between them. The API"},{"line_number":70,"context_line":"  checks this when the extra specs of a type with ``per_volume_qos_support``"},{"line_number":71,"context_line":"  are set. On other types these scoped keys are ignored, as other scoped"},{"line_number":72,"context_line":"  extra specs are."}],"source_content_type":"text/x-rst","patch_set":8,"id":"dd2a3892_9c65f19b","line":69,"in_reply_to":"b38db872_2f181b0c","updated":"2026-10-30 14:34:57.000000000","message":"Done. Create and os-set_qos are the authoritative check.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":75,"context_line":""},{"line_number":76,"context_line":"* For each limit, Cinder computes the value to apply: the per-volume value,"},{"line_number":77,"context_line":"  else the type default, else the range maximum. If the type has no range for"},{"line_number":78,"context_line":"  that limit, Cinder passes no value and the driver keeps applying the QoS"},{"line_number":79,"context_line":"  spec as today."},{"line_number":80,"context_line":"* Type defaults are not copied into the volume. A changed default applies the"},{"line_number":81,"context_line":"  next time the limits of the volume are applied."}],"source_content_type":"text/x-rst","patch_set":8,"id":"e34761a4_9d84cdb8","line":78,"updated":"2026-10-15 15:04:41.000000000","message":"\"else the range maximum\" means that adding a range to a type silently caps every volume of that type at the range maximum the next time QoS is applied, replacing whatever the QoS spec set. It also makes \"no per-volume limit, QoS spec applies as today\" impossible to express once a range exists.\n\nPlease make the chain: per-volume value, else type default, else existing QoS-spec behaviour (no computed value passed). The range should bound what users may set, not supply a value on its own.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":75,"context_line":""},{"line_number":76,"context_line":"* For each limit, Cinder computes the value to apply: the per-volume value,"},{"line_number":77,"context_line":"  else the type default, else the range maximum. If the type has no range for"},{"line_number":78,"context_line":"  that limit, Cinder passes no value and the driver keeps applying the QoS"},{"line_number":79,"context_line":"  spec as today."},{"line_number":80,"context_line":"* Type defaults are not copied into the volume. A changed default applies the"},{"line_number":81,"context_line":"  next time the limits of the volume are applied."}],"source_content_type":"text/x-rst","patch_set":8,"id":"6ad620e2_881e0230","line":78,"in_reply_to":"e34761a4_9d84cdb8","updated":"2026-10-30 14:34:57.000000000","message":"Agreed. Removed the range maximum fallback. The chain is now per-volume value, type default, then the QoS spec as today.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":79,"context_line":"  spec as today."},{"line_number":80,"context_line":"* Type defaults are not copied into the volume. A changed default applies the"},{"line_number":81,"context_line":"  next time the limits of the volume are applied."},{"line_number":82,"context_line":"* The driver maps the computed values to its own keys (for example"},{"line_number":83,"context_line":"  ``maxIOPS`` or ``total_iops_sec``), in place of the equivalent keys of the"},{"line_number":84,"context_line":"  QoS spec. Other QoS spec keys, such as separate read and write limits, still"},{"line_number":85,"context_line":"  apply."}],"source_content_type":"text/x-rst","patch_set":8,"id":"c45eceb8_e1130186","line":82,"updated":"2026-10-15 15:04:41.000000000","message":"\"In place of the equivalent keys of the QoS spec\" needs a rule a driver author can apply. Back-end QoS specs are vendor dictionaries: FlashArray has `maxIOPS`, `maxBWS`, `maxIOPS_per_GB`, `maxBWS_per_GB`; NetApp has absolute, per-GiB and adaptive-policy keys; PowerFlex has `maxIOPS`/`maxBWS`.\n\nSuggested rule: when a per-volume value exists for a dimension, the driver ignores every QoS-spec key that sets that dimension, absolute or per-capacity, and must keep honouring the per-volume value on paths that recompute per-capacity QoS (notably `extend_volume`). Please also decide the combined-vs-split case: if the spec carries separate read and write limits and the volume sets a combined max_iops, does the combined value win, do both apply (lower wins), or is that type invalid for per-volume QoS?","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":79,"context_line":"  spec as today."},{"line_number":80,"context_line":"* Type defaults are not copied into the volume. A changed default applies the"},{"line_number":81,"context_line":"  next time the limits of the volume are applied."},{"line_number":82,"context_line":"* The driver maps the computed values to its own keys (for example"},{"line_number":83,"context_line":"  ``maxIOPS`` or ``total_iops_sec``), in place of the equivalent keys of the"},{"line_number":84,"context_line":"  QoS spec. Other QoS spec keys, such as separate read and write limits, still"},{"line_number":85,"context_line":"  apply."}],"source_content_type":"text/x-rst","patch_set":8,"id":"ebd1f68b_3a85a139","line":82,"in_reply_to":"c45eceb8_e1130186","updated":"2026-10-30 14:34:57.000000000","message":"Done, using your rule, including extend. Split read/write limits still apply with a combined value, so the lower one wins.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":91,"context_line":"* The API rejects ``qos`` with 400 when the volume type does not have"},{"line_number":92,"context_line":"  ``per_volume_qos_support``, and when a value is outside the range of the"},{"line_number":93,"context_line":"  type. A limit without a range cannot be set per volume. This applies to"},{"line_number":94,"context_line":"  administrators too: to allow a value outside the range, they change the"},{"line_number":95,"context_line":"  range of the volume type."},{"line_number":96,"context_line":"* If an administrator narrows the range of a type, values already set on"},{"line_number":97,"context_line":"  volumes are kept. The new range is checked on the next ``os-set_qos``."}],"source_content_type":"text/x-rst","patch_set":8,"id":"27275e93_0503db74","line":94,"updated":"2026-10-15 15:04:41.000000000","message":"Binding administrators to the same range is a defensible choice, but please record it as a decision: say that an admin-only bypass rule (something like `volume:set_qos:unrestricted`) was considered and rejected, and why, so it is not re-opened during implementation.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":91,"context_line":"* The API rejects ``qos`` with 400 when the volume type does not have"},{"line_number":92,"context_line":"  ``per_volume_qos_support``, and when a value is outside the range of the"},{"line_number":93,"context_line":"  type. A limit without a range cannot be set per volume. This applies to"},{"line_number":94,"context_line":"  administrators too: to allow a value outside the range, they change the"},{"line_number":95,"context_line":"  range of the volume type."},{"line_number":96,"context_line":"* If an administrator narrows the range of a type, values already set on"},{"line_number":97,"context_line":"  volumes are kept. The new range is checked on the next ``os-set_qos``."}],"source_content_type":"text/x-rst","patch_set":8,"id":"b40daab8_328bb0b7","line":94,"in_reply_to":"27275e93_0503db74","updated":"2026-10-30 14:34:57.000000000","message":"Done. Added to Alternatives: admins can widen the range instead.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":98,"context_line":""},{"line_number":99,"context_line":"Volume operations:"},{"line_number":100,"context_line":""},{"line_number":101,"context_line":"* Create: for a type with ``per_volume_qos_support``, the volume manager calls"},{"line_number":102,"context_line":"  ``update_volume_qos`` after the driver created the volume, so drivers do not"},{"line_number":103,"context_line":"  need to change ``create_volume`` and its variants. If this fails, the volume"},{"line_number":104,"context_line":"  is set to ``error`` and a user message is created, the same as when a driver"}],"source_content_type":"text/x-rst","patch_set":8,"id":"ebc057b6_1872ae96","line":101,"updated":"2026-10-15 15:04:41.000000000","message":"Please state that `update_volume_qos` runs inside the create flow before the volume becomes available; otherwise there is a window where the volume is usable with no limit applied.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":98,"context_line":""},{"line_number":99,"context_line":"Volume operations:"},{"line_number":100,"context_line":""},{"line_number":101,"context_line":"* Create: for a type with ``per_volume_qos_support``, the volume manager calls"},{"line_number":102,"context_line":"  ``update_volume_qos`` after the driver created the volume, so drivers do not"},{"line_number":103,"context_line":"  need to change ``create_volume`` and its variants. If this fails, the volume"},{"line_number":104,"context_line":"  is set to ``error`` and a user message is created, the same as when a driver"}],"source_content_type":"text/x-rst","patch_set":8,"id":"155a677c_29f3a1d0","line":101,"in_reply_to":"ebc057b6_1872ae96","updated":"2026-10-30 14:34:57.000000000","message":"Done.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":109,"context_line":"  on success. On failure a user message is created and the stored values stay"},{"line_number":110,"context_line":"  unchanged. ``null`` clears a per-volume value, so the type default applies"},{"line_number":111,"context_line":"  again."},{"line_number":112,"context_line":"* Retype: per-volume values are cleared, because they were checked against"},{"line_number":113,"context_line":"  the old type, and the QoS settings of the new type are used. If the new type"},{"line_number":114,"context_line":"  has ``per_volume_qos_support``, its defaults apply, and the user can change"},{"line_number":115,"context_line":"  them later with ``os-set_qos``. If not, no per-volume limits are applied and"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7770de23_aca54dcb","line":112,"updated":"2026-10-15 15:04:41.000000000","message":"Clearing per-volume values on every retype drops a user\u0027s tuned limits without warning, including a retype between two types that both support per-volume QoS (for example, turning replication on). Suggest: keep the values when they fall within the new type\u0027s range; when they do not, either reject the retype with a clear error or clear them and emit a user message. Clearing on retype to a type without `per_volume_qos_support` is fine but should also emit a user message.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":109,"context_line":"  on success. On failure a user message is created and the stored values stay"},{"line_number":110,"context_line":"  unchanged. ``null`` clears a per-volume value, so the type default applies"},{"line_number":111,"context_line":"  again."},{"line_number":112,"context_line":"* Retype: per-volume values are cleared, because they were checked against"},{"line_number":113,"context_line":"  the old type, and the QoS settings of the new type are used. If the new type"},{"line_number":114,"context_line":"  has ``per_volume_qos_support``, its defaults apply, and the user can change"},{"line_number":115,"context_line":"  them later with ``os-set_qos``. If not, no per-volume limits are applied and"}],"source_content_type":"text/x-rst","patch_set":8,"id":"b730ced2_c3217159","line":112,"in_reply_to":"7770de23_aca54dcb","updated":"2026-10-30 14:34:57.000000000","message":"Done. Values within the new type\u0027s range are kept. Otherwise they are cleared with a user message, also when the new type has no support.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":114,"context_line":"  has ``per_volume_qos_support``, its defaults apply, and the user can change"},{"line_number":115,"context_line":"  them later with ``os-set_qos``. If not, no per-volume limits are applied and"},{"line_number":116,"context_line":"  the QoS spec of the new type applies as today."},{"line_number":117,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":118,"context_line":"  them."},{"line_number":119,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the limits."},{"line_number":120,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"c31c32b2_586fc29f","line":117,"updated":"2026-10-15 15:04:41.000000000","message":"Worth noting how this works: generic migration copies all non-skipped volumes columns into the temporary volume, so the new columns travel automatically and the destination\u0027s create path applies them. The destination must report `per_volume_qos_support; please say the scheduler enforces that on migrate as well as retype (line 62 implies it, but migration via `os-migrate_volume` with `force_host_copy` bypasses the filter).","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":114,"context_line":"  has ``per_volume_qos_support``, its defaults apply, and the user can change"},{"line_number":115,"context_line":"  them later with ``os-set_qos``. If not, no per-volume limits are applied and"},{"line_number":116,"context_line":"  the QoS spec of the new type applies as today."},{"line_number":117,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":118,"context_line":"  them."},{"line_number":119,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the limits."},{"line_number":120,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"03b8689a_cd09e4a6","line":117,"in_reply_to":"c31c32b2_586fc29f","updated":"2026-10-30 14:34:57.000000000","message":"Done. Migration to a backend without support is rejected, also with a given host.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":116,"context_line":"  the QoS spec of the new type applies as today."},{"line_number":117,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":118,"context_line":"  them."},{"line_number":119,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the limits."},{"line_number":120,"context_line":""},{"line_number":121,"context_line":"Driver contract:"},{"line_number":122,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"2de00aec_25cf57eb","line":119,"updated":"2026-10-15 15:04:41.000000000","message":"Under the rules above these volumes still receive the type\u0027s default (or, today, the range maximum), so \"do not inherit the limits\" is misleading. Suggest \"do not inherit the per-volume override; the type\u0027s defaults apply\".","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":116,"context_line":"  the QoS spec of the new type applies as today."},{"line_number":117,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":118,"context_line":"  them."},{"line_number":119,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the limits."},{"line_number":120,"context_line":""},{"line_number":121,"context_line":"Driver contract:"},{"line_number":122,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"2a3f1015_dbb406fb","line":119,"in_reply_to":"2de00aec_25cf57eb","updated":"2026-10-30 14:34:57.000000000","message":"Done, with your wording.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":120,"context_line":""},{"line_number":121,"context_line":"Driver contract:"},{"line_number":122,"context_line":""},{"line_number":123,"context_line":"* New optional driver method ``update_volume_qos(context, volume, qos)``,"},{"line_number":124,"context_line":"  called with the computed values. The base driver raises"},{"line_number":125,"context_line":"  ``NotImplementedError``. A driver that implements it reports"},{"line_number":126,"context_line":"  ``per_volume_qos_support\u003dTrue``."}],"source_content_type":"text/x-rst","patch_set":8,"id":"5ba74b17_5fbf9ea1","line":123,"updated":"2026-10-15 15:04:41.000000000","message":"Drivers need a defined way to refuse a value for a particular volume. Backends have minimum granularity (FlashArray: 100 IOPS, 1 MB/s; `_min` defaulting to 1 will be rejected by the array) and exclusivity rules (a FlashArray volume in a volume group that carries group QoS cannot also carry volume QoS; NetApp adaptive policies have no single maximum). Please specify an exception `update_volume_qos` may raise that the API maps to 400 on `os-set_qos` and to a user message on create, and allow drivers to publish minimums as capabilities (for example `per_volume_qos_min_iops`) so the API can validate before touching the backend.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"56675f88371e3c72104ce9721f7cdef55691cd34","unresolved":false,"context_lines":[{"line_number":120,"context_line":""},{"line_number":121,"context_line":"Driver contract:"},{"line_number":122,"context_line":""},{"line_number":123,"context_line":"* New optional driver method ``update_volume_qos(context, volume, qos)``,"},{"line_number":124,"context_line":"  called with the computed values. The base driver raises"},{"line_number":125,"context_line":"  ``NotImplementedError``. A driver that implements it reports"},{"line_number":126,"context_line":"  ``per_volume_qos_support\u003dTrue``."}],"source_content_type":"text/x-rst","patch_set":8,"id":"7909eab6_8ea42aee","line":123,"in_reply_to":"390e8f97_df99d9b4","updated":"2026-11-17 09:02:09.000000000","message":"Done.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":true,"context_lines":[{"line_number":120,"context_line":""},{"line_number":121,"context_line":"Driver contract:"},{"line_number":122,"context_line":""},{"line_number":123,"context_line":"* New optional driver method ``update_volume_qos(context, volume, qos)``,"},{"line_number":124,"context_line":"  called with the computed values. The base driver raises"},{"line_number":125,"context_line":"  ``NotImplementedError``. A driver that implements it reports"},{"line_number":126,"context_line":"  ``per_volume_qos_support\u003dTrue``."}],"source_content_type":"text/x-rst","patch_set":8,"id":"390e8f97_df99d9b4","line":123,"in_reply_to":"5ba74b17_5fbf9ea1","updated":"2026-10-30 14:34:57.000000000","message":"os-set_qos is async, so it can\u0027t return 400 after the driver call. Drivers report minimums as capabilities, the manager checks them, and drivers raise InvalidVolumeQoS otherwise. Both give a user message.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":128,"context_line":"  on attach or retype, not only in ``update_volume_qos``. Otherwise the volume"},{"line_number":129,"context_line":"  type QoS spec would replace them. A common helper returns the computed values"},{"line_number":130,"context_line":"  for a volume."},{"line_number":131,"context_line":"* For a replicated volume, the driver also applies the limits to the"},{"line_number":132,"context_line":"  secondary, so they stay after failover."},{"line_number":133,"context_line":""},{"line_number":134,"context_line":"NetApp ONTAP and Dell PowerFlex are proposed as the first implementations."}],"source_content_type":"text/x-rst","patch_set":8,"id":"4651ec3e_60878b6c","line":131,"updated":"2026-10-15 15:04:41.000000000","message":"\"The driver also applies the limits to the secondary\" is only possible where the backend replicates QoS with the volume. Synchronous active/active (for example an ActiveCluster pod) does; an async target holds a snapshot, not a volume, until failover. Suggest: \"where the backend supports it; otherwise the driver applies them when the secondary becomes active\".","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":128,"context_line":"  on attach or retype, not only in ``update_volume_qos``. Otherwise the volume"},{"line_number":129,"context_line":"  type QoS spec would replace them. A common helper returns the computed values"},{"line_number":130,"context_line":"  for a volume."},{"line_number":131,"context_line":"* For a replicated volume, the driver also applies the limits to the"},{"line_number":132,"context_line":"  secondary, so they stay after failover."},{"line_number":133,"context_line":""},{"line_number":134,"context_line":"NetApp ONTAP and Dell PowerFlex are proposed as the first implementations."}],"source_content_type":"text/x-rst","patch_set":8,"id":"3e7af8fb_500adc27","line":131,"in_reply_to":"4651ec3e_60878b6c","updated":"2026-10-30 14:34:57.000000000","message":"Done, with your wording.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":152,"context_line":"Data model impact"},{"line_number":153,"context_line":"-----------------"},{"line_number":154,"context_line":""},{"line_number":155,"context_line":"Two nullable columns on ``volumes``: ``qos_max_iops`` (Integer) and"},{"line_number":156,"context_line":"``qos_max_bps`` (BigInteger, since Integer stops at about 2.1 GB/s). ``NULL``"},{"line_number":157,"context_line":"means no per-volume value, so the rules above apply and existing rows need no"},{"line_number":158,"context_line":"data migration. The ``Volume`` object gets a new version, and"}],"source_content_type":"text/x-rst","patch_set":8,"id":"40adf2a9_f3e32280","line":155,"updated":"2026-10-15 15:04:41.000000000","message":"Two columns per key will not scale: burst and read/write split are explicitly deferred, and each would mean another migration and another `Volume` object version. A single nullable JSON column (`qos`) or a small `volume_qos` table keyed by volume id avoids that. If you keep the columns, please say why in the spec.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":152,"context_line":"Data model impact"},{"line_number":153,"context_line":"-----------------"},{"line_number":154,"context_line":""},{"line_number":155,"context_line":"Two nullable columns on ``volumes``: ``qos_max_iops`` (Integer) and"},{"line_number":156,"context_line":"``qos_max_bps`` (BigInteger, since Integer stops at about 2.1 GB/s). ``NULL``"},{"line_number":157,"context_line":"means no per-volume value, so the rules above apply and existing rows need no"},{"line_number":158,"context_line":"data migration. The ``Volume`` object gets a new version, and"}],"source_content_type":"text/x-rst","patch_set":8,"id":"0bbcd192_12117183","line":155,"in_reply_to":"40adf2a9_f3e32280","updated":"2026-10-30 14:34:57.000000000","message":"Agreed. Changed to one nullable JSON column `qos`, so new keys need no migration.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"d727959ea2ddc531e23f5131e56d21b7dc9c815e","unresolved":true,"context_lines":[{"line_number":293,"context_line":"References"},{"line_number":294,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":295,"context_line":""},{"line_number":296,"context_line":"None"}],"source_content_type":"text/x-rst","patch_set":8,"id":"ff52f2db_ae89b9eb","line":296,"updated":"2026-10-15 15:04:41.000000000","message":"Two specs interact with this one and should be referenced: the Xena user-visible extra specs spec (you rely on it at line 192) and the Pike capacity-based QoS spec (the per-GB keys this must override).","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"7f682c399c9fff2eb42afa74fec07aeb5a23ac13","unresolved":false,"context_lines":[{"line_number":293,"context_line":"References"},{"line_number":294,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":295,"context_line":""},{"line_number":296,"context_line":"None"}],"source_content_type":"text/x-rst","patch_set":8,"id":"5cf9728a_254722d3","line":296,"in_reply_to":"ff52f2db_ae89b9eb","updated":"2026-10-30 14:34:57.000000000","message":"Done.","commit_id":"847081d13dff361c2b2bb59b8e9dd571f86c1cda"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":53,"context_line":"Add two optional per-volume limits, ``max_iops`` and ``max_bps``, stored in"},{"line_number":54,"context_line":"a new ``qos`` column of the ``volumes`` table, not in volume metadata. Both"},{"line_number":55,"context_line":"are sustained upper limits for read and write combined. ``max_bps`` is an"},{"line_number":56,"context_line":"integer number of bytes per second, with no prefixes; drivers round it to the"},{"line_number":57,"context_line":"backend\u0027s granularity. Burst settings are out of scope: burst limits in the"},{"line_number":58,"context_line":"volume type QoS spec still apply unchanged. Drivers apply new limits to in-use"},{"line_number":59,"context_line":"volumes without requiring a detach or interrupting IO."}],"source_content_type":"text/x-rst","patch_set":10,"id":"afc31efd_cb7f333d","line":56,"updated":"2026-11-01 15:36:21.000000000","message":"A refinement of my own PS8 L47 wording: \"round it to the backend\u0027s granularity\" gives no direction, and both proposed first implementations round up today. NetApp REST converts bytes per second to MB/s with `math.ceil` (`netapp/dataontap/client/client_cmode_rest.py:1615`) and PowerFlex rounds the KB/s limit up to the next 1024 (`dell_emc/powerflex/driver.py:2131-2134`), so a value equal to the type\u0027s `_max` can be applied slightly above it. Please say \"round it down to the backend\u0027s granularity, so the applied limit never exceeds the requested value\"; the minimum side is covered by the `per_volume_qos_min_*` capabilities at 145-149. Optionally say after 220 that the stored and displayed value is the requested one.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":53,"context_line":"Add two optional per-volume limits, ``max_iops`` and ``max_bps``, stored in"},{"line_number":54,"context_line":"a new ``qos`` column of the ``volumes`` table, not in volume metadata. Both"},{"line_number":55,"context_line":"are sustained upper limits for read and write combined. ``max_bps`` is an"},{"line_number":56,"context_line":"integer number of bytes per second, with no prefixes; drivers round it to the"},{"line_number":57,"context_line":"backend\u0027s granularity. Burst settings are out of scope: burst limits in the"},{"line_number":58,"context_line":"volume type QoS spec still apply unchanged. Drivers apply new limits to in-use"},{"line_number":59,"context_line":"volumes without requiring a detach or interrupting IO."}],"source_content_type":"text/x-rst","patch_set":10,"id":"8f23604b_9004dd2f","line":56,"in_reply_to":"afc31efd_cb7f333d","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":99,"context_line":"API checks:"},{"line_number":100,"context_line":""},{"line_number":101,"context_line":"* A new microversion adds an optional ``qos`` object to volume create and a"},{"line_number":102,"context_line":"  volume action ``os-set_qos`` for ``available`` or ``in-use`` volumes."},{"line_number":103,"context_line":"* The API rejects ``qos`` with 400 when the volume type does not have"},{"line_number":104,"context_line":"  ``per_volume_qos_support``, and when a value is outside the range of the"},{"line_number":105,"context_line":"  type. A limit without a range cannot be set per volume. This applies to"}],"source_content_type":"text/x-rst","patch_set":10,"id":"781ed5c5_b5fe5989","line":102,"updated":"2026-11-01 15:36:21.000000000","message":"Status alone is not a sufficient precondition here or in the 400 list at 213-215. Retype (`cinder/volume/api.py:1956-1960`) and migrate (:1759-1760) also require `migration_status` in `AVAILABLE_MIGRATION_STATUS` (:108), because during a generic migration the volume stays available or in-use while `migration_status` is `migrating`. An `os-set_qos` accepted in that window is cast to the source host and its stored value is overwritten when `finish_volume_migration` copies the temporary volume\u0027s fields back (`cinder/objects/volume.py:526-565`), after the detail view and billing have already seen it. Nothing says what state the volume is in while the cast is in flight either, so two `os-set_qos calls`, or one and a retype, can reach the driver and the database in different orders, breaking 121-122. Please say the API claims the volume with a `conditional_update` on status and `migration_status` as retype does, then state one of: status unchanged with the manager serialising `os-set_qos` per volume under a coordination lock, or a transitional status while the driver call runs. Add \"migration or retype in progress\" to the 400 reasons at 213-215.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":99,"context_line":"API checks:"},{"line_number":100,"context_line":""},{"line_number":101,"context_line":"* A new microversion adds an optional ``qos`` object to volume create and a"},{"line_number":102,"context_line":"  volume action ``os-set_qos`` for ``available`` or ``in-use`` volumes."},{"line_number":103,"context_line":"* The API rejects ``qos`` with 400 when the volume type does not have"},{"line_number":104,"context_line":"  ``per_volume_qos_support``, and when a value is outside the range of the"},{"line_number":105,"context_line":"  type. A limit without a range cannot be set per volume. This applies to"}],"source_content_type":"text/x-rst","patch_set":10,"id":"176b2646_6573a6bc","line":102,"in_reply_to":"781ed5c5_b5fe5989","updated":"2026-11-17 09:01:40.000000000","message":"Done: conditional_update, then a per-volume lock.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":103,"context_line":"* The API rejects ``qos`` with 400 when the volume type does not have"},{"line_number":104,"context_line":"  ``per_volume_qos_support``, and when a value is outside the range of the"},{"line_number":105,"context_line":"  type. A limit without a range cannot be set per volume. This applies to"},{"line_number":106,"context_line":"  administrators too: to allow a value outside the range, they change the"},{"line_number":107,"context_line":"  range of the volume type."},{"line_number":108,"context_line":"* If an administrator narrows the range of a type, values already set on"},{"line_number":109,"context_line":"  volumes are kept. The new range is checked on the next ``os-set_qos``."}],"source_content_type":"text/x-rst","patch_set":10,"id":"b281b10c_65f8897c","line":106,"updated":"2026-11-01 15:36:21.000000000","message":"Blocker, and my apologies for not seeing it earlier. This sentence, with 91-92, 108-109, 171-172, 234-236 and 265-267, assumes an administrator can edit `per_volume_qos_support` and the `per_volume_qos:*` keys on a type that already has volumes. Cinder refuses that: `_allow_update()` at `cinder/api/contrib/types_extra_specs.py:62-70` raises 400 \"Volume Type is currently in use.\" as soon as one volume of the type exists, from create (:89), update (:118) and delete (:166). So once the first volume is created the range, the default and the support flag are frozen; \"admins can widen the range instead\" and \"narrows the range\" cannot happen, per-volume QoS cannot be switched on for an existing populated type, and the only route is a new type plus a retype of every volume, which is the burden lines 26-27 say this spec removes (retype to the same type is refused too, `cinder/volume/api.py:1960`). By contrast `qos_specs.update()` (`cinder/volume/qos_specs.py:70`) has no in-use check and association with an in-use type is allowed, which undercuts the reason given at 166-170 for not holding the range there.\n\nPlease choose one and write it down. (a) Hold the range and default in the QoS spec, the one carrier operators can change while volumes exist; Cinder strips the `per_volume_qos:` namespace in `volume_types.get_volume_type_qos_specs()` (`volume_types.py:341`) before drivers see it, which also answers the NetApp unknown-key concern (`netapp/utils.py:292-307`), and the type\u0027s user-visible view exposes the range read-only. (b) Exempt the `scoped per_volume_qos:*` keys from the in-use lock, keeping `per_volume_qos_support locked since it drives scheduling, and state what happens to stored values when a range shrinks. (c) Keep the lock and say plainly that a range is fixed once the type has a volume, rewriting every sentence above that tells admins to change it. My preference is (a); it also fits the alternative at 173.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":103,"context_line":"* The API rejects ``qos`` with 400 when the volume type does not have"},{"line_number":104,"context_line":"  ``per_volume_qos_support``, and when a value is outside the range of the"},{"line_number":105,"context_line":"  type. A limit without a range cannot be set per volume. This applies to"},{"line_number":106,"context_line":"  administrators too: to allow a value outside the range, they change the"},{"line_number":107,"context_line":"  range of the volume type."},{"line_number":108,"context_line":"* If an administrator narrows the range of a type, values already set on"},{"line_number":109,"context_line":"  volumes are kept. The new range is checked on the next ``os-set_qos``."}],"source_content_type":"text/x-rst","patch_set":10,"id":"33c1d7a7_f7a2a23d","line":106,"in_reply_to":"b281b10c_65f8897c","updated":"2026-11-17 09:01:40.000000000","message":"Done, option (a). The keys are also stripped in qos_specs.get_qos_specs(), which NetApp, PowerFlex and Pure read directly.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":110,"context_line":""},{"line_number":111,"context_line":"Volume operations:"},{"line_number":112,"context_line":""},{"line_number":113,"context_line":"* Create: for a type with ``per_volume_qos_support``, the volume manager calls"},{"line_number":114,"context_line":"  ``update_volume_qos`` in the create flow after the driver created the"},{"line_number":115,"context_line":"  volume and before the volume becomes ``available``, so drivers do not need"},{"line_number":116,"context_line":"  to change ``create_volume`` and its variants. If this fails, the volume"}],"source_content_type":"text/x-rst","patch_set":10,"id":"557cff3f_8dd38577","line":113,"updated":"2026-11-01 15:36:21.000000000","message":"Volumes created with a group from a group snapshot or a source group never enter this flow: `cinder/volume/api.py:344-355` passes no scheduler or volume RPC API when `group_snapshot` or `source_group` is set, and `manager.create_group_from_src` (:3433) calls the driver\u0027s group clone methods and `_update_volume_from_src` (:3706) with no `update_volume_qos` call and no user message on failure. So for a type with a `_default`, 132-133 promises a value that only a driver applying QoS inside its clone methods would deliver, which 116 says it need not. Please either say the manager makes the same call for each volume of `create_group_from_src` before the volumes become available and add it to 298, or say at 132-133 that these volumes receive the type default when the driver next applies QoS and qualify 116.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":110,"context_line":""},{"line_number":111,"context_line":"Volume operations:"},{"line_number":112,"context_line":""},{"line_number":113,"context_line":"* Create: for a type with ``per_volume_qos_support``, the volume manager calls"},{"line_number":114,"context_line":"  ``update_volume_qos`` in the create flow after the driver created the"},{"line_number":115,"context_line":"  volume and before the volume becomes ``available``, so drivers do not need"},{"line_number":116,"context_line":"  to change ``create_volume`` and its variants. If this fails, the volume"}],"source_content_type":"text/x-rst","patch_set":10,"id":"605ddb82_ed7e13f2","line":113,"in_reply_to":"557cff3f_8dd38577","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":115,"context_line":"  volume and before the volume becomes ``available``, so drivers do not need"},{"line_number":116,"context_line":"  to change ``create_volume`` and its variants. If this fails, the volume"},{"line_number":117,"context_line":"  is set to ``error`` and a user message is created, the same as when a driver"},{"line_number":118,"context_line":"  fails to apply the volume type QoS spec during ``create_volume``. The request"},{"line_number":119,"context_line":"  is not rescheduled, since the volume already exists on the backend. The"},{"line_number":120,"context_line":"  backend volume is removed when the user deletes the volume in ``error``."},{"line_number":121,"context_line":"* ``os-set_qos``: the driver is called first and the database is updated only"}],"source_content_type":"text/x-rst","patch_set":10,"id":"faaeb8d9_d76991ab","line":118,"updated":"2026-11-01 15:36:21.000000000","message":"Nit: only the user message is the same. A driver exception inside `create_volume` is rescheduled by default (`OnFailureRescheduleTask.revert`, `cinder/volume/flows/manager/create_volume.py:189-222`) and reaches error only when attempts are exhausted, whereas line 119 deliberately does not reschedule. Suggest: \"a user message is created, as for a create_volume failure, and the volume is set to error. Unlike a `create_volume` failure the request is not rescheduled, since the volume already exists on the backend.\"","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":115,"context_line":"  volume and before the volume becomes ``available``, so drivers do not need"},{"line_number":116,"context_line":"  to change ``create_volume`` and its variants. If this fails, the volume"},{"line_number":117,"context_line":"  is set to ``error`` and a user message is created, the same as when a driver"},{"line_number":118,"context_line":"  fails to apply the volume type QoS spec during ``create_volume``. The request"},{"line_number":119,"context_line":"  is not rescheduled, since the volume already exists on the backend. The"},{"line_number":120,"context_line":"  backend volume is removed when the user deletes the volume in ``error``."},{"line_number":121,"context_line":"* ``os-set_qos``: the driver is called first and the database is updated only"}],"source_content_type":"text/x-rst","patch_set":10,"id":"843975c2_a33883da","line":118,"in_reply_to":"faaeb8d9_d76991ab","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":122,"context_line":"  on success. On failure a user message is created and the stored values stay"},{"line_number":123,"context_line":"  unchanged. ``null`` clears a per-volume value, so the type default applies"},{"line_number":124,"context_line":"  again."},{"line_number":125,"context_line":"* Retype: per-volume values within the range of the new type are kept."},{"line_number":126,"context_line":"  Values outside it or without a range in the new type, or all values when"},{"line_number":127,"context_line":"  the new type has no ``per_volume_qos_support``, are cleared and a user"},{"line_number":128,"context_line":"  message is created."}],"source_content_type":"text/x-rst","patch_set":10,"id":"cb78e782_e859c901","line":125,"updated":"2026-11-01 15:36:21.000000000","message":"Still open from PS8 (my L112): the outcome is settled, but not when it happens, and the migration bullet at 129 contradicts it for a retype that migrates. On master `manager.retype` calls `driver.retype` only on the same backend (`cinder/volume/manager.py:3071` onwards) while the volume still carries the old type, and any change in `per_volume_qos:*` keys fails `_can_use_driver_migration` (:2648), so a cross-backend retype takes `_migrate_volume_generic` (:2362), which copies every non-skipped column, the new qos column included, into the temporary volume before anything could clear it. A helper keyed on the volume alone (143-144) also returns the old type\u0027s default inside `driver.retype`, where drivers apply QoS today with `new_type` (`pure.py:3597`). Please add that the kept values are computed from the new type before the retype or migration starts, that the temporary volume is created with the kept values only, that the stored values and the user message are written on success so a failed retype leaves them unchanged, and that the helper takes an optional volume type which the driver passes as `new_type`. Qualify 129 to migration without a type change.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":122,"context_line":"  on success. On failure a user message is created and the stored values stay"},{"line_number":123,"context_line":"  unchanged. ``null`` clears a per-volume value, so the type default applies"},{"line_number":124,"context_line":"  again."},{"line_number":125,"context_line":"* Retype: per-volume values within the range of the new type are kept."},{"line_number":126,"context_line":"  Values outside it or without a range in the new type, or all values when"},{"line_number":127,"context_line":"  the new type has no ``per_volume_qos_support``, are cleared and a user"},{"line_number":128,"context_line":"  message is created."}],"source_content_type":"text/x-rst","patch_set":10,"id":"ca46231b_495cccc7","line":125,"in_reply_to":"cb78e782_e859c901","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":126,"context_line":"  Values outside it or without a range in the new type, or all values when"},{"line_number":127,"context_line":"  the new type has no ``per_volume_qos_support``, are cleared and a user"},{"line_number":128,"context_line":"  message is created."},{"line_number":129,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":130,"context_line":"  them. Migration of a volume whose type has ``per_volume_qos_support`` to a"},{"line_number":131,"context_line":"  backend without it is rejected, also when a host is given."},{"line_number":132,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the"}],"source_content_type":"text/x-rst","patch_set":10,"id":"52442cf5_132880e0","line":129,"updated":"2026-11-01 15:36:21.000000000","message":"Nit: \"the destination backend applies them\" holds for `_migrate_volume_generic`, where the column copy and the destination create flow do the work, but driver-assisted migration has no destination step. `manager.migrate_volume` (`cinder/volume/manager.py:2670-2722`) calls `driver.migrate_volume` on the source and saves only host, cluster, status and AZ fields; the backend keeps the limits with the volume. Please say so: \"for driver-assisted migration the volume keeps its limits on the array; for generic migration the destination create flow applies them\".","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":126,"context_line":"  Values outside it or without a range in the new type, or all values when"},{"line_number":127,"context_line":"  the new type has no ``per_volume_qos_support``, are cleared and a user"},{"line_number":128,"context_line":"  message is created."},{"line_number":129,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":130,"context_line":"  them. Migration of a volume whose type has ``per_volume_qos_support`` to a"},{"line_number":131,"context_line":"  backend without it is rejected, also when a host is given."},{"line_number":132,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the"}],"source_content_type":"text/x-rst","patch_set":10,"id":"2ecd50fc_d58151d1","line":129,"in_reply_to":"52442cf5_132880e0","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":128,"context_line":"  message is created."},{"line_number":129,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":130,"context_line":"  them. Migration of a volume whose type has ``per_volume_qos_support`` to a"},{"line_number":131,"context_line":"  backend without it is rejected, also when a host is given."},{"line_number":132,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the"},{"line_number":133,"context_line":"  per-volume values; the defaults of their type apply."},{"line_number":134,"context_line":""}],"source_content_type":"text/x-rst","patch_set":10,"id":"f0eb2550_266193f3","line":131,"updated":"2026-11-01 15:36:21.000000000","message":"First a correction of my own PS8 premise: `force_host_copy` does not bypass the filter. `backend_passes_filters` (`cinder/scheduler/filter_scheduler.py:123`) runs the full filter pass against the named host, so no scheduler change is needed. But \"is rejected\" reads like the 400 at 103, and it is not: `os-migrate_volume` returns 202 (`cinder/volume/api.py:1777` only sets `migration_status` to `starting`) and the CapabilitiesFilter miss happens in the scheduler, which sets `migration_status` to `error` with no user message (`scheduler/manager.py:289-306`). Suggest: \"fails in the scheduler like any other CapabilitiesFilter miss (`migration_status` error), also when a host is given\". If you want a user message for parity with the retype bullet, say so and add it to Work Items.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":128,"context_line":"  message is created."},{"line_number":129,"context_line":"* Migration keeps the per-volume values, and the destination backend applies"},{"line_number":130,"context_line":"  them. Migration of a volume whose type has ``per_volume_qos_support`` to a"},{"line_number":131,"context_line":"  backend without it is rejected, also when a host is given."},{"line_number":132,"context_line":"* Volumes created from a snapshot, volume or backup do not inherit the"},{"line_number":133,"context_line":"  per-volume values; the defaults of their type apply."},{"line_number":134,"context_line":""}],"source_content_type":"text/x-rst","patch_set":10,"id":"76695c7d_b1da177c","line":131,"in_reply_to":"f0eb2550_266193f3","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":143,"context_line":"  type QoS spec would replace them. A common helper returns the computed values"},{"line_number":144,"context_line":"  for a volume."},{"line_number":145,"context_line":"* Drivers may report ``per_volume_qos_min_iops`` and"},{"line_number":146,"context_line":"  ``per_volume_qos_min_bps``. The volume manager checks them before calling the"},{"line_number":147,"context_line":"  driver. For other cases, such as group QoS on the backend, the driver raises"},{"line_number":148,"context_line":"  ``InvalidVolumeQoS``. Both result in a user message, and the stored values"},{"line_number":149,"context_line":"  stay unchanged."}],"source_content_type":"text/x-rst","patch_set":10,"id":"54370356_b2e59063","line":146,"updated":"2026-11-01 15:36:21.000000000","message":"Still open from PS8 (my L123). I accept that an RPC cast cannot return 400, and the capabilities plus `InvalidVolumeQoS` are what I asked for. But \"before calling the driver\" does not say whether on create the check runs before c`reate_volume` or after it, and 113-120 frame every create-time QoS failure as leaving a backend volume behind. The manager already holds `self.driver.capabilities` before the create flow starts (`manager.py:887`), so a value below the reported minimum can be refused without creating anything; please say so, and say whether that refusal reschedules or errors with a user message (the latter is simpler). In Other deployer impact please add that `_min` should be set at or above the minimum of every backend behind the type, because the default of 1 is below FlashArray\u0027s floor of 100 IOPS and 1 MB/s (`pure.py:194-196`).","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":143,"context_line":"  type QoS spec would replace them. A common helper returns the computed values"},{"line_number":144,"context_line":"  for a volume."},{"line_number":145,"context_line":"* Drivers may report ``per_volume_qos_min_iops`` and"},{"line_number":146,"context_line":"  ``per_volume_qos_min_bps``. The volume manager checks them before calling the"},{"line_number":147,"context_line":"  driver. For other cases, such as group QoS on the backend, the driver raises"},{"line_number":148,"context_line":"  ``InvalidVolumeQoS``. Both result in a user message, and the stored values"},{"line_number":149,"context_line":"  stay unchanged."}],"source_content_type":"text/x-rst","patch_set":10,"id":"8ca43405_bc7a3c3e","line":146,"in_reply_to":"54370356_b2e59063","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":170,"context_line":"  need an empty one only to hold the range."},{"line_number":171,"context_line":"* An admin-only rule to set values outside the range: admins can widen the"},{"line_number":172,"context_line":"  range instead, and a single rule keeps behaviour and billing predictable."},{"line_number":173,"context_line":"* QoS types separate from volume types, as Manila added in Gazpacho [1]_. In"},{"line_number":174,"context_line":"  Manila, a QoS type is still chosen through the share type"},{"line_number":175,"context_line":"  (``default_qos_type``), so all shares of a type get the same QoS, and there"},{"line_number":176,"context_line":"  is no API to change the QoS of an existing share. Manila states that changing"}],"source_content_type":"text/x-rst","patch_set":10,"id":"2ab75b8e_5c505389","line":173,"updated":"2026-11-01 15:36:21.000000000","message":"This rebuts Manila\u0027s QoS types as shipped, and the description is accurate against manila master. The obvious Cinder extension of that model is not recorded, though: a nullable per-volume reference to an existing QoS spec (`volumes.qos_specs_id`, the shape of Manila\u0027s `share_instances.qos_type_id`), with the type listing the QoS specs it permits and `os-set_qos` taking a name. Cinder already has the pieces (`VolumeType.qos_specs_id`, the association API in `cinder/volume/qos_specs.py`, `qos_specs` in the create `request_spec`, the retype qos diff), and QoS specs are editable in use while type extra specs are not, so someone is likely to raise it on Brian\u0027s open thread. The spec does answer it in substance: use cases 1, 3, 4 and 6 need continuous values, one curated QoS spec per combination reproduces the type explosion of 24-26, and 166-167 already notes project members cannot read QoS specs. Please add two sentences here saying exactly that, so the rejection covers the extension explicitly. If you take option (a) at 106 the two bullets become one argument.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":170,"context_line":"  need an empty one only to hold the range."},{"line_number":171,"context_line":"* An admin-only rule to set values outside the range: admins can widen the"},{"line_number":172,"context_line":"  range instead, and a single rule keeps behaviour and billing predictable."},{"line_number":173,"context_line":"* QoS types separate from volume types, as Manila added in Gazpacho [1]_. In"},{"line_number":174,"context_line":"  Manila, a QoS type is still chosen through the share type"},{"line_number":175,"context_line":"  (``default_qos_type``), so all shares of a type get the same QoS, and there"},{"line_number":176,"context_line":"  is no API to change the QoS of an existing share. Manila states that changing"}],"source_content_type":"text/x-rst","patch_set":10,"id":"cf8f1183_92744685","line":173,"in_reply_to":"2ab75b8e_5c505389","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":181,"context_line":"Data model impact"},{"line_number":182,"context_line":"-----------------"},{"line_number":183,"context_line":""},{"line_number":184,"context_line":"One nullable JSON column ``qos`` on ``volumes``, for example"},{"line_number":185,"context_line":"``{\"max_iops\": 3000, \"max_bps\": 125000000}``. New keys, such as read and write"},{"line_number":186,"context_line":"limits later, need no new column. ``NULL`` or a missing key means no"},{"line_number":187,"context_line":"per-volume value, so existing rows need no data migration. The ``Volume``"}],"source_content_type":"text/x-rst","patch_set":10,"id":"fba0d044_5359e800","line":184,"updated":"2026-11-01 15:36:21.000000000","message":"Clarification: since \"JSON column\" was my phrase on PS8. Cinder has no native JSON column type: `cinder/db/models.py` uses `sa.Text` for its two dict-valued columns, `volume_attachment.connection_info` and `connector` (:498, :500), decoded by the object onto a `DictOfNullableField`. Suggest: \"One nullable Text column `qos` on `volumes` holding a JSON-encoded object, decoded by the Volume object onto a `DictOfNullableField`, the same pattern as `volume_attachment.connection_info`.\" The NULL or missing-key sentence is fine.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":181,"context_line":"Data model impact"},{"line_number":182,"context_line":"-----------------"},{"line_number":183,"context_line":""},{"line_number":184,"context_line":"One nullable JSON column ``qos`` on ``volumes``, for example"},{"line_number":185,"context_line":"``{\"max_iops\": 3000, \"max_bps\": 125000000}``. New keys, such as read and write"},{"line_number":186,"context_line":"limits later, need no new column. ``NULL`` or a missing key means no"},{"line_number":187,"context_line":"per-volume value, so existing rows need no data migration. The ``Volume``"}],"source_content_type":"text/x-rst","patch_set":10,"id":"eba36b2b_03b0c542","line":184,"in_reply_to":"fba0d044_5359e800","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":231,"context_line":"Security impact"},{"line_number":232,"context_line":"---------------"},{"line_number":233,"context_line":""},{"line_number":234,"context_line":"Project members can change the limits of their own volumes, but only within"},{"line_number":235,"context_line":"the range set by the administrator in the volume type. Administrators are"},{"line_number":236,"context_line":"bound by the same range, and only they can change it."},{"line_number":237,"context_line":""}],"source_content_type":"text/x-rst","patch_set":10,"id":"89e0f190_d9d98847","line":234,"updated":"2026-11-01 15:36:21.000000000","message":"Rajat\u0027s L39 thread is still open and this is the paragraph a reviewer with that concern reads, but it only restates the range bound. The answer exists in the spec and in your Gerrit reply: nothing is exposed unless an administrator opts a type in with `per_volume_qos_support` and a range (103-105); the default of `volume:set_qos` is the same as `volume:create` (`cinder/policies/volumes.py:74-75`) and a deployment can restrict it to administrators; and the worst case is every volume of the type at `_max`, the same exposure as a QoS spec that sets `_max` today, since Cinder has no aggregate IOPS or bandwidth quota. Please add those sentences here so the thread can be closed on the text. Whether the range can change once the type has volumes is my comment at 106, not this one.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":231,"context_line":"Security impact"},{"line_number":232,"context_line":"---------------"},{"line_number":233,"context_line":""},{"line_number":234,"context_line":"Project members can change the limits of their own volumes, but only within"},{"line_number":235,"context_line":"the range set by the administrator in the volume type. Administrators are"},{"line_number":236,"context_line":"bound by the same range, and only they can change it."},{"line_number":237,"context_line":""}],"source_content_type":"text/x-rst","patch_set":10,"id":"b4216c5a_0e9e9be4","line":234,"in_reply_to":"89e0f190_d9d98847","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":243,"context_line":"Notifications impact"},{"line_number":244,"context_line":"--------------------"},{"line_number":245,"context_line":""},{"line_number":246,"context_line":"``os-set_qos`` sends ``volume.set_qos.start`` and ``volume.set_qos.end``"},{"line_number":247,"context_line":"notifications with the new values, so billing systems can track them."},{"line_number":248,"context_line":""},{"line_number":249,"context_line":"Other end user impact"}],"source_content_type":"text/x-rst","patch_set":10,"id":"85a375bd_147f2a9b","line":246,"updated":"2026-11-01 15:36:21.000000000","message":"The billing use case at 46-48 needs the current limits at every billing event, not only on `os-set_qos`. Volume notifications are built by `_usage_from_volume` (`cinder/volume/volume_utils.py:93`), a fixed dict with no `qos` key plus caller-supplied extra info, so `volume.create.end` for a volume created with `qos`, `volume.retype` after a retype that keeps or clears values, and `volume.exists` from the usage audit would all omit the limits. The metadata workaround this spec replaces was at least visible to billing through `volume_metadata` in the payload. Please add one sentence: the `qos` values are added to the volume usage payload, so create, retype, exists and the set_qos notifications all carry the current limits.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":243,"context_line":"Notifications impact"},{"line_number":244,"context_line":"--------------------"},{"line_number":245,"context_line":""},{"line_number":246,"context_line":"``os-set_qos`` sends ``volume.set_qos.start`` and ``volume.set_qos.end``"},{"line_number":247,"context_line":"notifications with the new values, so billing systems can track them."},{"line_number":248,"context_line":""},{"line_number":249,"context_line":"Other end user impact"}],"source_content_type":"text/x-rst","patch_set":10,"id":"9b15a2e6_a9e9c285","line":246,"in_reply_to":"85a375bd_147f2a9b","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":13425,"name":"Simon Dodsley","email":"simon@everpuredata.com","username":"sdodsley"},"change_message_id":"453f2e58b7c2eabacdf2bc2808b55ebbc89fedb5","unresolved":true,"context_lines":[{"line_number":303,"context_line":"Dependencies"},{"line_number":304,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":305,"context_line":""},{"line_number":306,"context_line":"None. Builds on user visible extra specs [2]_ and capacity-based QoS [3]_."},{"line_number":307,"context_line":""},{"line_number":308,"context_line":"Testing"},{"line_number":309,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":10,"id":"a49e97cd_e823a84e","line":306,"updated":"2026-11-01 15:36:21.000000000","message":"Also my imprecision on PS8, sorry. The Pike spec defines only the front-end `*_iops_sec_per_gb` keys, which the volume manager scales into connection_info for the hypervisor (`_parse_connection_options`, `cinder/volume/manager.py:1849-1882`); drivers never see them, so \"builds on capacity-based QoS\" contradicts the back-end-only scope at 61-64. The per-capacity keys the override rule at 93-97 must handle are vendor back-end keys (Pure `maxIOPS_per_GB`, NetApp `maxIOPSperGiB`, PowerFlex `maxIOPSperGB`). Suggest: \"None. Builds on user visible extra specs [2]. *The per-capacity keys covered by the override rule are vendor back-end QoS spec keys; the front-end per-GB keys of capacity-based QoS [3]* are applied by the hypervisor and are not changed here.\" Keep [3] in References.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"},{"author":{"_account_id":36686,"name":"Inyong Hong","display_name":"hongp","email":"inyong.hong@samsung.com","username":"hong-p"},"change_message_id":"13026f4bf61e46638a46f78e64b9b04ee5370b59","unresolved":false,"context_lines":[{"line_number":303,"context_line":"Dependencies"},{"line_number":304,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":305,"context_line":""},{"line_number":306,"context_line":"None. Builds on user visible extra specs [2]_ and capacity-based QoS [3]_."},{"line_number":307,"context_line":""},{"line_number":308,"context_line":"Testing"},{"line_number":309,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":10,"id":"3961aa73_c84fa2b4","line":306,"in_reply_to":"a49e97cd_e823a84e","updated":"2026-11-17 09:01:40.000000000","message":"Done.","commit_id":"7029fb3c2b5412a970aee9f224c149ad6867cf70"}]}
