<fix>[ha]: ZSTAC-84992 direct-pick HA pre-fence core to 5.4.10 (@@3)#4512
Open
MatheMatrix wants to merge 42 commits into
Open
<fix>[ha]: ZSTAC-84992 direct-pick HA pre-fence core to 5.4.10 (@@3)#4512MatheMatrix wants to merge 42 commits into
MatheMatrix wants to merge 42 commits into
Conversation
DBImpact Resolves: ZSTAC-81425 Change-Id: I627a6576686f6c6972636d67786e77716e637471
<fix>[conf]: bump version to 5.4.10 See merge request zstackio/zstack!9045
Use entity alias for simple return-with-total counts so nullable selected fields do not turn totals into zero. Keep explicit count(field) and distinct(field) behavior unchanged. Add CountGroupByCase coverage for nullable first selected field. Resolves: ZSTAC-78110 Change-Id: I3b437bc686c345ba17baf256a338ce88209571f8 (cherry picked from commit 0dfe389)
The vhost-user interface (used by OVS-DPDK) strictly requires the 'virtio-net' frontend driver in Libvirt/QEMU. However, ZStack defaults Windows NIC drivers to 'e1000' (or other emulated models) when the image platform is 'Windows', causing VM start failure with Libvirt error: "vhostuser requires the virtio-net* frontend". This fix forces the NIC driver to 'virtio' specifically when the interface type is 'vhostuser' on Windows platforms, regardless of the default platform settings. Satisfying both the legacy boot requirements of Windows and the vhost-user network requirements. Resolves: ZSTAC-76013 Change-Id: I737a6e637273796d776e726e7969786c7a6a6478
Route AutoQuery handling through the query sync queue so bursty GraphQL refreshes cannot run unlimited APIQueryMessage requests concurrently. Resolves: ZSTAC-80742 Change-Id: If2c052e539526cd0511e037d1718838bdb65a7a3
<fix>[kvm]: force virtio driver for windows vhostuser nic See merge request zstackio/zstack!10342
Backport the normalizedModelName schema required by premium GPU spec deduplication to 5.4.10. This is the zstack companion change for the premium ZSTAC-86476 backport. Constraint: ZSTAC-86476 is the 5.4.10 clone of ZSTAC-75319. Confidence: high Scope-risk: narrow Tested: ./runMavenProfile premium in pr-build-bin Docker with zstack and premium mounted Not-tested: local stability case execution Resolves: ZSTAC-86476 Change-Id: Ic77e24e974a99842c40f29e4bd72851e3833795b
Backport ZSTAC-74748 pagination tiebreaker from 5.5.6/5.5.28 to 5.4.10. Paginated LIMIT/OFFSET queries can return duplicate resources when the requested sort field has duplicate values. Append the primary key as a deterministic tiebreaker in both ZQL and legacy HQL query paths, and skip hidden PK fields that ZQL cannot order by. Source-Commits: 284d049, b1156d8 Tested: git diff --check upstream/5.4.10..HEAD Resolves: ZSTAC-86475 Change-Id: I14bef1d630a70711ef2605ea95fd37cb86e5273c
Existing host reconnect can fail before kvmagent echo. Ansible masks libvirt sockets with systemctl during deploy. If host systemd D-Bus is stuck, that optional step times out. Continue only for this known mask timeout on reconnect. New host deploy and other ansible failures still fail. Resolves: ZSTAC-86349 Change-Id: I0ef4a535065ff797c9e4cfae5b39c2daa321a4cc
Resolves: ZSTAC-86350 Change-Id: I024ae51da6e5784fb336d8fbd5a4aeab44500565
Backport the management node hash ring fix to 5.4.10. 1. Why? Management node lifecycle reconciliation can race with callbacks. This can cause NodeHash or Nodes inconsistency and task stalling. 2. How? Synchronize hash ring operations and use defensive copies. Add lifecycle reconciliation locking and two-round missing checks. 3. Side effects? Management node lifecycle operations are serialized by a local lock. # Summary of changes (by module): - core: synchronize resource destination hash ring operations. - portal: serialize management node lifecycle reconciliation. Source-Commit: 66ab572 Related: ZSTAC-86294, ZSTAC-78182 Change-Id: I3ba21ffc7b8f1284e17d50da7b027ffdad70828b
Backport external-storage split-brain guards to 5.4.10. 1. Why? Starting a VM on a new host may continue even when the old storage client still holds the volume. For xinfini Vhost volumes this can leave two QEMU processes using the same bdev. 2. How? Report xinfini deactivate failures instead of swallowing them, verify the bdev is really deleted, re-check active clients after deactivate success, and make blacklist failures propagate out of beforeStartVmOnKvm. 3. Side effects? Deactivate failure or a still-active old client now aborts VM start for that volume instead of silently continuing. Happy-path deactivate keeps the previous behavior. # Summary of changes (by module): - header: make PrimaryStorageNodeSvc.blacklist synchronous. - externalStorage: post-check active clients and propagate blacklist failures. - xinfini: surface deactivate/delete failures and abort unsupported isolation. - expon/zbs: adapt blacklist implementations to the synchronous signature. Related: ZSTAC-86444 Related: ZSTAC-85124 Change-Id: I59812c6d68eec6533e48a8c9cf05b7c55d6b85eb
Backport VM HA skip-trace protection to 5.4.10. 1. Why? When a management node leaves, its VM skip-trace entries are removed immediately. If those VMs are still being started by kvmagent, the next VM sync may treat them as stopped and trigger HA, causing split-brain. 2. How? Move departed management node skip entries into an orphaned set, keep them skipped for a configurable TTL, include them in VM sync skip lists, and remove expired entries with compare-and-remove cleanup. 3. Side effects? VMs whose controller MN departed stay skipped until the TTL expires. The default TTL is 600 seconds and can be changed by global config `kvm.vm.orphanedSkipTimeout`. # Summary of changes (by module): - kvm: add orphaned skip set, TTL global config, sync inclusion and CAS cleanup. Related: ZSTAC-84636 Related: ZSTAC-80821 Change-Id: I7862360c7318cedb51e38939467cf2a7836e61d0
[5.4.10][ZSTAC-86350] <fix>[identity]: restrict APIQueryAccountMsg to admin-only in RBAC See merge request zstackio/zstack!10380
Backport PMU-off handling from ZSTAC-76375 to 5.4.10 for Kunpeng openEuler guest panic. Resolves: ZSTAC-86338 Change-Id: I3c2dbe1944c300a803f50198b4c7ba575bd77dcc
<fix>[kvm]: skip known systemd timeout on reconnect See merge request zstackio/zstack!10362
Resolves: ZSTAC-78989 Change-Id: I0fe3a56ab724978944c69afadaab7ff7353e4c0f (cherry picked from commit f19223a)
Pass dns-forward-max and cache-size through VR DNS commands. Render both options in zstack-vyos dnsmasq config. Resolves: ZSTAC-85197 Change-Id: Ib78504edc564a2ab7dbd39bcb2e0ee396d019240 (cherry picked from commit 4ada5a5)
Change-Id: If07dce918c4d7d9f62445184ff5984e30b83ce64 (cherry picked from commit 8939994)
Trigger TF port sync from TF controller ping success instead of MN startup. Keep the original one day interval and port cleanup algorithm, while scoping sync to the current controller. Resolves: ZSTAC-85281 Change-Id: I84a88a34135cd47fc79b3019d2f0fc2953f06ac9 (cherry picked from commit e44d41a)
Remove DHCP message handlers and generated error code reference unavailable on 5.4.10. Change-Id: I1a15fb1eeeab60ca9c6fa239904bda89839b64e5
Add MessageSafe to the SDN message entrypoint. Move the second ping after the first async sync window in the focused test. Resolves: ZSTAC-86445 Change-Id: I4d7748a0de3b5d6a052a47966d2265249b2779e2
Resolves: ZSTAC-79353 Change-Id: I63706d727173706e796b64666963776a74736c61 (cherry picked from commit 2ef761d)
Resolves: ZSTAC-81450 Change-Id: I79666f616d64797a747677656762626d6a71786d (cherry picked from commit 5c3fab5)
[ZSTAC-86394] Backport ZSTAC-78110 to 5.4.10 See merge request zstackio/zstack!10335
[ZSTAC-86484] Backport ZSTAC-78989 to 5.4.10 See merge request zstackio/zstack!10337
[ZSTAC-86501] Backport ZSTAC-79353 to 5.4.10 See merge request zstackio/zstack!10338
[ZSTAC-86489] Backport ZSTAC-85197 to 5.4.10 See merge request zstackio/zstack!10341
[ZSTAC-86445] Backport ZSTAC-85281 to 5.4.10 See merge request zstackio/zstack!10340
<fix>[kvm]: disable arm vm pmu by default See merge request zstackio/zstack!10382
[ZSTAC-86470] Backport ZSTAC-81450 to 5.4.10 See merge request zstackio/zstack!10336
<fix>[search]: stabilize paginated query order See merge request zstackio/zstack!10375
<fix>[gpu]: add gpu normalized model schema See merge request zstackio/zstack!10370
The premium backport adds minAvailableDevicesPerHost to GetMdevDeviceCandidates. Commit the generated SDK action field on the companion zstack branch so update-sdk CI no longer reports an uncommitted SDK diff. Constraint: premium and zstack companion branches must use the same @@2 suffix for cross-repo SDK validation Tested: git diff --check Not-tested: Full ./runMavenProfile sdk locally Resolves: ZSTAC-86497 Change-Id: Iae56b01a4c533eeba05d10358d4944f655c5a6d4
ZSTAC-84636 backport VM HA orphaned skip-trace fix See merge request zstackio/zstack!10401
ZSTAC-86444 backport external-storage split-brain guard See merge request zstackio/zstack!10399
<fix>[mn]: synchronize hash ring operations See merge request zstackio/zstack!10394
[5.4.10] cherry-pick ZSTAC query fix See merge request zstackio/zstack!10343
<fix>[sdk]: update mdev candidates sdk See merge request zstackio/zstack!10420
Move premium-only HA pre-fence details out of community while keeping the generic async before-start extension point support. Jira: ZSTAC-83890 Test: mvn -pl header,compute,plugin/kvm,simulator/simulatorImpl,testlib -DskipTests compile Change-Id: I168adf82338f9df9e76287619b7f76a8e5be695f
Move HA pre-fence pending tag creation into HaStartVmInstanceMsg handling after the VM state update, so later pre-fence execution observes the same database state. Resolves: ZSTAC-83890 Change-Id: I7f046732e0245319ed4cf188d1342b65dad11767
Create HA pre-fence pending tag only when the checker provides an accessible peer host. The peer should be covered by the HA dead-host and storage-inactive paths, so do not persist a fallback marker when it is absent. Resolves: ZSTAC-83890 Change-Id: I36426b3386f1080ddeae8396ea9c4f7a81e25711
Owner
Author
|
Comment from gitlab: 检测到该PR状态为cannot_be_merged, 所以移除了REVIEWED标签 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Direct cherry-pick the ZSTAC-83890 zstack commits to 5.4.10 for ZSTAC-84992 without squashing or rewriting commit messages.
This MR uses the unified @@3 branch name required across zstack, premium, and zstack-utility:
cp/5.4.10/ZSTAC-84992-direct-pick-20260709@@3Supersedes previous direct-pick MR without @@3 branch suffix: http://dev.zstack.io:9080/zstackio/zstack/-/merge_requests/10472
Supersedes previous squash/backport MR: http://dev.zstack.io:9080/zstackio/zstack/-/merge_requests/10457
Cherry-picked commits
<fix>[ha]: stabilize suspect host pre-fence flow<fix>[ha]: create pre-fence tag in core<fix>[ha]: require peer for pre-fence tagValidation
git diff --checkResolves: ZSTAC-84992
sync from gitlab !10473