voonhous opened a new issue, #19306:
URL: https://github.com/apache/hudi/issues/19306
Tracking issue for a full triage of the open PR backlog. All 555 open PRs as
of 2026-07-16, from #2607 (2021-02-26) to #19305 (2026-07-16), were reviewed
against master at e098da78c501 (2026-07-15) to decide what can be closed, what
should be converted into an issue, and what is still viable.
The full per-PR tables, with a one-line reason for every verdict, are in the
first three comments below:
- Comment 1: CLOSE (166)
- Comment 2: CONVERT to issue, then close (166)
- Comment 3: KEEP (223)
These verdicts are recommendations, not decisions. If you are the author or
a reviewer of a PR listed under CLOSE or CONVERT and believe it should stay
open, please comment here or on the PR.
## Scope and method
- Covers every open apache/hudi PR as of 2026-07-16, from #2607 (2021-02-26)
to
#19305 (2026-07-16).
- Every PR was assessed against master at e098da78c501 (2026-07-15): whether
the
touched files still exist, whether an equivalent fix or feature already
landed
(the landing PR/commit is named in the reason where found), and whether the
subsystem has since been rewritten (1.0 timeline/FileGroupReader rewrites,
hive-sync rewrite, removed engine version poms, etc.).
- Branch policy used: release-0.x/1.0.x and branch-0.x are frozen -- if the
fix
also applies to master the verdict is CONVERT, otherwise CLOSE (active
committer backports were kept case-by-case). release-1.2.0 is the active
upcoming release branch and was judged on activity. asf-site PRs were
judged
on whether the docs change is still accurate for the current site.
feature-branch-rfc100-unstructured-data is treated as active.
- Recent PRs (roughly May-Jul 2026) in normal review flow default to KEEP;
[DNM] tags were read in context (stacking/merge-gate markers vs abandoned
experiments).
## Verdicts
- CLOSE (166): obsolete -- already landed or superseded on master, code
removed or rewritten, dead target branch, or abandoned after blocking review
feedback.
- CONVERT (166): the underlying bug/feature is still relevant on current
master, but the PR is too stale to rescue. File a GitHub issue capturing the
scope, then close the PR.
- KEEP (223): still viable -- applies (or nearly applies) to master, or has
active author/committer engagement; worth reviewing or rebasing as-is.
## Summary by year created
| Year | CLOSE | CONVERT | KEEP | Total |
|---|---|---|---|---|
| 2021 | 4 | 1 | 0 | 5 |
| 2022 | 25 | 39 | 1 | 65 |
| 2023 | 22 | 42 | 6 | 70 |
| 2024 | 31 | 24 | 5 | 60 |
| 2025 | 54 | 46 | 31 | 131 |
| 2026 | 30 | 14 | 180 | 224 |
| Total | 166 | 166 | 223 | 555 |
## Borderline calls flagged during triage
These verdicts were judgment calls; a maintainer close to the area should
double-check
before acting.
From the oldest 400:
- #3417: the only 2021-vintage PR that still merges cleanly, but zero review
traction since 2021 (bot bumps only); left as CONVERT rather than KEEP.
- #4665 (thrift-mode hive sync): CLOSE because HoodieHiveClient was removed
in the hive-sync rewrite; arguable CONVERT if thrift mode is still wanted.
- #5758 vs #18445 (confluent bump): master is still on confluent 5.5.0, so
the upgrade is genuinely outstanding; the 2022 PR is CONVERT, the 2026 one
(#18445) is KEEP.
- #6337, #6409, #6448, #6858: could not fully confirm
landed-vs-still-broken; marked CONVERT so scope is not lost, verify before
filing issues.
- #7680 (show/alter table properties): 3 approvals, but alter-props
partially landed via AlterTableCommand; CONVERT covers only the remaining "show
properties" scope.
- #7748 (RFC-42 bucket-resizing amendment): doc-only change on a merged RFC;
CONVERT, but CLOSE is defensible.
- #8673 (Flink wrong-instant race): the coordinator was reworked with
pending-instant recommit logic, so the race may already be fixed; verify before
converting.
- #9188 (index perf fix): draft, but committer-updated 2025-12, hence KEEP.
- #9554: unclear whether the spark-schema-evolution-then-hive-read bug
survives the refactored hadoop-mr path; CONVERT conservatively.
- #10308, #10396, #10406: niche / stacked-on-unmerged /
semi-legacy-bootstrap; all CONVERT but each is a defensible CLOSE.
- #11047: approved cosmetic rename, 2 years stale; CLOSE, though a
maintainer could rebase it.
- #11573 (Java 17 CI): CONVERT -- bot.yml is still JDK 11.
- #11706 (SI rollback/compaction test draft): CLOSE unless that coverage gap
is confirmed still open.
- #12608 (UnMerged log refactor): CLOSE as an abandoned pure-refactor with
16 unaddressed review rounds; CONVERT if the refactor is still desired.
- #12732 (record positions from Spark SQL): core position-write infra landed
via #9581; the remaining SQL-DML gap may be partly covered already.
- #12747 (merger rename): KEEP because yihua rebased it in 2026, but CI is
failing and rename refactors rot fast.
- #12824 (branch-0.x timestamp ordering): approved for 0.x; CLOSE assumes no
further 0.x release is planned.
- #12982: reached an approval before being abandoned on a since-refactored
FSView subsystem; CLOSE.
- #13348 (MDT key-lookup perf): CLOSE as superseded-by-own-author rework,
but the perf goal itself is arguably still open.
- #13499 (RFC-97 payload deprecation doc): KEEP, though the deprecation work
is landing on master without it.
- #13582, #13745: small approved-or-applied changes stalled most of a year;
KEEP vs CONVERT is a coin flip.
- #13954, #14041, #14051 (test-coverage PRs): CONVERT to preserve coverage
intent; given the project's test-bloat concerns, plain CLOSE is defensible.
- #14154, #14156, #14160, #14193 (Lance RFC-100 stack): all CONFLICTING and
mostly idle since late 2025; kept because
feature-branch-rfc100-unstructured-data is active and they depend on the
unmerged reader PR #14137.
- #14248 (Flink schema-on-write): large and CONFLICTING, dormant since Dec
2025, but active committer author; KEEP.
- #17692: trivial CI image bump by a PMC member; called CLOSE as stale,
could still be wanted.
- #17796 (XL MDT test/benchmark framework): heavy review investment but
stalled ~5 months; KEEP, could be CONVERT if dropped.
- #18159 (blob cleaner): KEEP, but the RFC-100 plan was restructured by
merged #18013 and cleaner work continues on a feature branch, so it may
effectively be superseded.
- #18247 (aircompressor 0.27 -> 2.0.3 CVE bump): KEEP since the CVE is
unaddressed, but the major-version jump likely needs manual API migration.
- #18338 (in-repo Claude Code skills): mergeable and active; whether
.claude/ belongs in-repo is a maintainer call.
- #18346 (RFC-90 re-post): zero engagement; CONVERT borders on CLOSE.
- #18350, #18384: recent committer PRs buried under 90-100 file conflicts;
KEEP, but CONVERT is reasonable.
From the newest 155:
- #18504 (RFC-103 LSM phase 1): 3 months stale and conflicting, but blocked
on a design question the author posed to the maintainers -- ball is in their
court; KEEP.
- #18655 (SchemaChangeLogger): closest CONVERT-vs-KEEP call in the tail;
stale unreviewed draft, scope still relevant.
- #18795 (RollbackOrphanDetector): KEEP, but maintainers could not reproduce
the underlying issue, so it may end up rejected.
- #18806, #18807 vs #18810 (test-only repro PRs): split on engagement -- the
first two have committer review (KEEP), #18810 has none plus red CI (CONVERT).
- #19138, #19228, #19247, #19083 ([DNM]-tagged): DNM read as
stacking/merge-gate markers on active work, not abandonment; all KEEP. #19229
by contrast was DNM onto frozen release-0.14.1 with zero reviews (CLOSE).
- #19248, #19305 (branch-0.x): kept despite the freeze rule -- branch-0.x
still merged fixes in Apr 2026 and both are active committer backports; a
master port of #19248's fix may also be needed.
- #19270-#19273: stacked asf-site blog series with identical create
timestamps; all four kept as active drafts rather than treating later ones as
superseding earlier.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
