feat(auto-increment): add transactional epoch fence#26027
Conversation
Co-authored-by: ULookup <CHBulookup@outlook.com>
…uto-increment-epoch-fence
Co-authored-by: ULookup <CHBulookup@outlook.com>
Co-authored-by: ULookup <CHBulookup@outlook.com>
…uto-increment-epoch-fence
Qodo reviews are paused for this user.Troubleshooting steps vary by plan Learn more → On a Teams plan? Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center? |
XuPeng-SH
left a comment
XuPeng-SH
left a comment
There was a problem hiding this comment.
Approved at exact head a7510bf.
Correctness review:
- The dedicated epoch is persisted independently of ordinary schema version and TN enforces monotonic current+1 transitions with overflow rejection.
- CN propagation is closed across in-memory writes, object writes, deletes, soft deletes, transfer, workspace flush, and CN-object compaction. Flush grouping includes table+epoch+known, so generations cannot be coalesced accidentally.
- The live race is closed by TAE prepare ordering: DML prepared first publishes its prepare watermark before a later ALTER validates; ALTER prepared first publishes a committing schema node that later DML waits on and then rejects by epoch.
- Legacy unknown writers map to epoch zero only, so they remain compatible before the first reset and fail closed afterward.
- WAL replay obtains the real dirty-table set from the serialized TxnMemo; replayed prepared transactions fence the table until both commit and rollback terminal paths resolve them.
Unhappy-path review found no blocking leak, hang, stale-fence, or unbounded-growth path. The committed watermark is monotonic; prepared replay registrations have symmetric commit/rollback cleanup; schema waits terminate through the transaction state transition.
Fresh local validation at this exact head passed both normal and race-focused matrices for disttae, catalog, rpc, txnimpl, and TAE db tests. git diff --check also passed. Current GitHub CI is green.
Non-blocking requirements for the follow-up user-facing integration:
- The allocator reset/cache protocol must guarantee that the epoch attached to a write is the generation that actually allocated its values; merely reading a refreshed relation epoch after allocation would be insufficient.
- Enabling reset must be gated against old TNs that do not enforce these new protobuf fields.
- COPY/CLONE reconciliation and end-to-end restart/multi-CN coverage should remain explicit acceptance gates, as stated in #23143.
Process nit: the PR body still says intentionally Draft, while the PR is currently Ready for review.
Merge Queue Status
This pull request spent 22 minutes 41 seconds in the queue, with no time running CI. Waiting for
All conditions
ReasonPull request #26027 has been dequeued Pull request from fork cannot be queued. This pull request comes from a fork, and Mergify needs the author's permission to update its branch.
HintYou should look at the reason for the failure and decide if the pull request needs to be fixed or if you want to requeue it. Tick the box to put this pull request back in the merge queue (same as
|
4 participants
What type of PR is this?
Which issue(s) this PR fixes:
issue #23143
What this PR does / why we need it:
This is the first AUTO_INCREMENT-specific prerequisite split from the closed #25883. It does not expose
ALTER TABLE ... AUTO_INCREMENT=Nyet.The invariant introduced here is: a CN write carries the AUTO_INCREMENT epoch observed when its values were allocated, and TN accepts the write only when that epoch matches the table epoch. Once a table enters a fenced epoch, legacy writers that cannot provide an epoch fail closed.
SchemaExtra/plan.TableDef, propagate it through disttae workspace, row and object writes, flush/compaction, RPC, and TN relationscurrent + 1Additional guards reject epoch exhaustion before mutation and keep ordinary schema ALTER operations from advancing the AUTO_INCREMENT epoch.
Explicit non-goals
ALTER TABLE ... AUTO_INCREMENT=Nplanner or compile entry point;Diff size
+1283/-131;+1565/-1079;Validation
go test ./pkg/vm/engine/disttae ./pkg/vm/engine/tae/catalog ./pkg/vm/engine/tae/rpc ./pkg/vm/engine/tae/txn/txnimpl ./pkg/vm/engine/tae/db/test-racecoverage for epoch transitions and replay commit/rollback lifecyclegit diff --checkmo-pr-preflight-review: PASS at exact head4122a7add078c79757121a4add2d8771113c7980mo-self-review: no open blockerThe PR is intentionally Draft while its full CI runs and while this prerequisite is reviewed independently.