docs: alpha5 release notes cover the seven time-series features

[jdatcmd]

The alpha5 release notes before the tag. They were drafted on 2026-10-12, before the owner made the seven time-series features the gate for alpha5, and named encoding as the only theme. All seven are now merged (#1292 closed with #1330), so the notes describe them.

@OffgridwithJD, please check the facts against what you built, above all tiering, partitions and the daemon expire.

What changes in RELEASE_NOTES_1.0-alpha5.md

• Introduction and highlights: two themes, time series and encoding. A highlight for the seven features, and one for the changed default.
• A new "Time series" section, one subsection per feature:
  • daemon expire (1.0: the maintenance daemon should run expire() for tables that declare a TTL #1290, with expire reads a column's zone maps in one catalog scan, not one per row group #1310);
  • partitions (1.0: partition lifecycle for time-partitioned columnar tables, via pg_partman #1291, Native partition lifecycle for time-partitioned columnar tables: create ahead, drop past retention #1297's create-ahead);
  • Parquet tiering (1.0: tier cold partitions to Parquet on object storage #1295);
  • incremental rollups (1.0: incremental rollups (continuous aggregates) over columnar tables #1294);
  • bucketed queries (1.0: is GROUP BY date_bin(...) vectorized? Measure, then make it so #1292, 1.0: answer max(ts) and latest-per-device queries from zone maps #1293, opt: estimate time-bucket groups from the zone maps, and recognise date_bin (#1292) #1304);
  • gap filling and approximate aggregates (1.0: gap filling, LOCF and approximate aggregates: recipes, compatibility, or build #1296).
• Settings: pgcolumnar.autovacuum_expire is added, and pgcolumnar.enable_group_vectorization changes from off to on. The two developer settings, rollup_pause and tier_fault_after, get one sentence. I diffed the setting definitions in columnar_tableam.c between v1.0-alpha4 and main: those five are the only new ones, and none was removed.
• Fixed: the collation fix (Zone maps keep the old collation after ALTER COLUMN ... COLLATE, so queries return wrong results #1327), the COMMIT PREPARED crash (COMMIT PREPARED in the preparing session crashes the backend after a pgcolumnar write #1323), and the unlocked partition open on a non-recursing ALTER TABLE.
• Known issues: the cap error that a plan made under a higher cap can still hit (Grouped vectorized aggregate: spill or hand off to core past groupagg_max_groups, instead of erroring #1300), and rollup refusals near retained history.
• What the upgrade changes: the four new catalogs, the new functions, and column_extent starting empty for existing tables.

How the statements were checked

• Sources: each statement comes from the feature's CHANGELOG entry or its reference section. I corrected four that my first draft overstated:
  • a float8 average in a rollup can differ in the last digits;
  • max(ts)/min(ts) read no data only for groups without deletes;
  • lock_wait bounds every lock tier_partition takes;
  • an existing table's extent appears at its next recompute, not after enough writes.
• Style: docs_style 58/58.

Also done for the tag, not in this PR

• Upgrade from v1.0-alpha4: extension_upgrade.sh v1.0-alpha4 passes on PG15–19 at main b55cb23. The upgraded catalog is byte-identical to a fresh alpha5 install (211 objects) on each major.

Still to do on tag day

The release date (TODO(release)), renaming the CHANGELOG's [Unreleased] section, and the tag.

🤖 Generated with Claude Code

[OffgridwithJD]

Review at 7c30706. I checked the release notes against the source rather than against the CHANGELOG.

Checked against the source, all correct

• Settings: I diffed the Define*Variable calls between v1.0-alpha4 and main. Exactly five are new (autovacuum_expire, enable_post_codec_encoding_choice, index_min_blocks, rollup_pause, tier_fault_after), and none was removed.
  • rollup_pause and tier_fault_after are PGC_SUSET, so "only a superuser can set them" is right.
  • autovacuum_expire is PGC_SIGHUP, so "takes effect on reload" is right.
• Catalogs: I diffed the CREATE TABLEs in the two install scripts. Exactly four are new: column_extent, retired_group, rollup and rollup_group.
• Functions: I diffed the CREATE FUNCTIONs. The seven named in "What the upgrade changes", plus rollup_source_groups and rollup_source_retired as "two internal helpers". Nothing else.
• What I built:
  • The tiering paragraph matches #1317: ACCESS EXCLUSIVE from the detach to commit, every lock bounded by lock_wait, and the foreign partition keeping its name, bounds, grants and comment.
  • The collation fix and the COMMIT PREPARED crash read right.
  • The known issue on the cap is accurate now that #1330 gates on a bound.

Three changes

1. #1331 isn't in it yet. The highlight says "pg_partman and two new functions manage time partitions", but the "Time partitions" section and the upgrade list name only partition_create_ahead. Once #1331 merges, these need:

   • partition_drop_plan in that section;
   • partition_drop_plan and its helper partition_drop_candidates in "What the upgrade changes", which makes it two helpers for rollups and one for partitions.

   A sentence for the section, if it helps:

   pgcolumnar.partition_drop_plan(parent, cutoff) returns the statements that drop every partition older than the cutoff: a concurrent detach and a drop each, run with \gexec. It never takes an exclusive lock on the parent, and a drop refuses while its partition is still attached.

2. The rollup known issue misses the refusal #1328's review added. A refresh also refuses (0A000) when rows were deleted from a row group that expire then retired before the refresh. The deleted rows are gone, so their buckets can't be recomputed. Suggested:

   A refresh also refuses when rows were deleted from a row group that expire retired before the refresh.

3. #1310, one word stronger than the code. "When nothing can qualify, it answers from the column's extent" is true only when a complete extent shows it. A table whose extent isn't complete yet still reads every group's zone map. Suggested:

   When the column's extent shows that nothing can qualify, it answers in one index lookup instead of reading every row group's zone map.

docs_style I took from your run. I haven't rerun it.

[jdatcmd]

@OffgridwithJD, thank you for checking against the source. All three are in at 6a5a958:

1. partition_drop_plan: in "Time partitions", using your sentence, and in the upgrade list. That list now ends "three internal helpers: two for rollups and one for partitions". I checked feat: pgcolumnar.partition_drop_plan returns the statements that retire old time partitions (#1297) #1331's upgrade script: it adds partition_drop_candidates and the two partition_drop_plan overloads, nothing else.
2. The delete-then-expire refusal is in the rollup known issue, with your wording.
3. expire reads a column's zone maps in one catalog scan, not one per row group #1310: "When the column's extent shows that nothing can qualify, it answers in one index lookup…"

docs_style 58/58 at this head.

This PR merges after #1331, so the notes never name a function main doesn't have. #1331's merge is waiting on CI.

[jdatcmd]

@OffgridwithJD, the release notes at 6a5a958 carry your three corrections, and CI is 15/15. With #1331 merged and #1334 about to merge, this is the last content PR before the tag-day PR (release date and CHANGELOG rename). Please review when you can.

Also, for the tag: at the owner's request I rebuilt my container's PostgreSQL 15–19 from fresh release tarballs, with contrib, hll, tdigest and pg_partman on every major. A full matrix and pytest run on all five majors is under way at 614eb02. I'll post the results here before asking the owner for the tag.

[OffgridwithJD]

Approved at 6a5a958.

• All three corrections are in.
• partition_drop_plan is in both the Time partitions section and the upgrade list.
• Diffing the CREATE FUNCTIONs between v1.0-alpha4 and main gives exactly the eight named plus three helpers: rollup_source_groups, rollup_source_retired and partition_drop_candidates.
• The settings and catalog claims are unchanged from my earlier check, and partition_drop_plan added neither.

[jdatcmd]

Merging: approved by @OffgridwithJD at 6a5a958, and CI is green on that head.