MDEV-40467 [to-squash] Improve Y2038/2106 problem error message

[mariadb-YuchenPei]

When nearing the end times TIMESTAMP_MAX_VALUE, auto-partition creation may encounter overflow when calculating new partition range. This patch improves the message of such an overflow

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[AITWINS]

Armageddon tenet.

…

On Wednesday, July 22, 2026, Yuchen Pei ***@***.***> wrote: When nearing the end times TIMESTAMP_MAX_VALUE, auto-partition creation may encounter overflow when calculating new partition range. This patch improves the message of such an overflow ------------------------------ You can view, comment on, or merge this pull request online at: #5436 Commit Summary - 13d3076 <13d3076> MDEV-40467 [to-squash] Improve Y2038/2106 problem error message File Changes (3 files <https://github.com/MariaDB/server/pull/5436/files>) - *M* mysql-test/main/partition_range_interval.result <https://github.com/MariaDB/server/pull/5436/files#diff-44c74af4aa93edc2588bde835a1996987fe90d939d545bd2a90bb51e3341e0c2> (67) - *M* mysql-test/main/partition_range_interval.test <https://github.com/MariaDB/server/pull/5436/files#diff-145158c0f9da292fdea568e379966f678f6bd9cff039089e9b5ac57ab45b5c78> (43) - *M* sql/sql_partition_auto_interval.cc <https://github.com/MariaDB/server/pull/5436/files#diff-a4d7d34ba7762e95ec7a170d94cea2a77759bd8687a5b25cafbeb5366792961c> (8) Patch Links: - https://github.com/MariaDB/server/pull/5436.patch - https://github.com/MariaDB/server/pull/5436.diff — Reply to this email directly, view it on GitHub <#5436?email_source=notifications&email_token=BFASTMKXKDFKVGZJOABAVFL5GA2GFA5CNFSNUABEM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UF42DCMBUHE3TMMBUGGTHEZLBONXW5KTTOVRHGY3SNFRGKZFFMV3GK3TUVRTG633UMVZF6Y3MNFRWW>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BFASTMIBR6XXFNWQYIFAQT35GA2GFAVCNFSNUABEKJSXA33TNF2G64TZHMYTSOBRGYYDOMB3JFZXG5LFHM2DSNBVGIZDGMJSHCQXMAQ> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/BFASTMNMZLX37D444YZCLZD5GA2GFA5CNFSNUABEM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UF42DCMBUHE3TMMBUGGTHEZLBONXW5KTTOVRHGY3SNFRGKZFFMV3GK3TUVJTG633UMVZF62LPOM> and Android <https://github.com/notifications/mobile/android/BFASTMOPUMOKT4ANWGHVL3D5GA2GFA5CNFSNUABEM5UWIORPF5TWS5BNNB2WEL2QOVWGYUTFOF2WK43UF42DCMBUHE3TMMBUGGTHEZLBONXW5KTTOVRHGY3SNFRGKZFFMV3GK3TUVZTG633UMVZF6YLOMRZG62LE>. Download it today! You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[mariadb-YuchenPei]

claude review comment

---

Review: 13d30768 — MDEV-40467 [to-squash] Improve Y2038/2106 problem error message

Overview

Same targeted code change as the prior revision, now with a complete, self-consistent test. In check_range_interval_constants(), a TIMESTAMP partition boundary that crosses TIMESTAMP_MAX_VALUE during DML-triggered auto-partition creation raises a clean ER_DATA_OUT_OF_RANGE. The test (at timestamp=4294967295, i.e. 2106-02-07) covers TIMESTAMP (errors) plus DATE and DATETIME (succeed), each with recorded show create table output.

Resolved from previous reviews ✅

• Type-scoping fix is correct — the if (error) guard lives inside case MYSQL_TYPE_TIMESTAMP:, so DATE/DATETIME columns aren't spuriously rejected near 2106.
• .test/.result are now in sync — the previously-missing show create table t1; output is recorded for all three cases. The TIMESTAMP case correctly shows only p0 (proving the failed auto-creation is atomic — no partial p1…pN are added on error), while DATE/DATETIME show p0…p7 created up to the 2106-02-08 boundary. This is a genuinely good, behavior-locking test.

Correctness

• Logic is sound. The error out-param from Timestamp_or_zero_datetime is only consulted where it's meaningful, and the recorded results match the expected partition math (transition 2106-02-01 + 1-day steps up to end = query_start, one overshoot to 2106-02-08).

Potential risk — timezone determinism (worth a look)

• The number of partitions created depends on converting the max timestamp (4294967295) to local time via the session time zone (gmt_sec_to_TIME(&end_time, thd->query_start())), and the recorded result hard-codes p0…p7 (last boundary 2106-02-08). The test does not pin time_zone. Under a sufficiently negative UTC offset, end could fall on 2106-02-06, yielding one fewer partition and a result mismatch.
  • In practice MTR runs with a fixed default TZ so CI is likely stable, but relying on that is implicit. Recommend making it explicit and self-documenting:

    SET time_zone='+00:00';
    ...
    SET time_zone=default;

  This also guards against future MTR/host changes.

Minor

• Duplicated my_error block — now identical to the date_add_interval overflow handler earlier in the same function. Optional: extract a tiny helper. Not blocking.
• set timestamp never restored — add set timestamp=default; at the end per convention (it's the last block today, so harmless, but tidy).
• Comment nit: with the check now scoped to TIMESTAMP, /* Likely Year 2038/2106 Problem */ could drop "Likely" — the cause is definite (timestamp range exceeded).

Performance & security

• No performance impact (single branch on the cold partition-creation path); no security concerns.

Summary

This revision resolves both prior findings — the code change is correct and well-scoped, and the test is now consistent and meaningfully verifies TIMESTAMP-errors / DATE-DATETIME-succeed plus error atomicity. The only substantive follow-up is pinning time_zone so the recorded partition count is deterministic across environments; the duplicated my_error and missing set timestamp=default; are cosmetic. Good to squash once the time-zone point is addressed (or consciously accepted).

[mariadb-YuchenPei]

claude:

---

Review: 2e423a48 — MDEV-40467 [to-squash] Improve Y2038/2106 problem error message

Overview

Same code change as prior revisions, with the test now pinned to a fixed time zone. In check_range_interval_constants(), a TIMESTAMP auto-interval partition boundary crossing TIMESTAMP_MAX_VALUE raises a clean ER_DATA_OUT_OF_RANGE. The test (time_zone='+00:00', timestamp=4294967295) covers TIMESTAMP (errors, only p0 remains) plus DATE and DATETIME (succeed, p0…p7 created).

All prior findings resolved ✅

• Type-scoping — if (error) correctly lives inside case MYSQL_TYPE_TIMESTAMP:; DATE/DATETIME unaffected.
• .test/.result sync — show create table output recorded for all three cases; TIMESTAMP case confirms error atomicity (no partial p1…pN).
• Time-zone determinism — SET time_zone='+00:00' pins the max-timestamp→local conversion, so the recorded p0…p7 count is stable across environments. With +00:00, end = 2106-02-07 06:28:15 and stepping from 2106-02-01 yields exactly p0…p7 — matches the recorded result. Correctly restored with SET time_zone=default;.

Correctness

• Logic and recorded results are consistent and reproducible. No issues.

Minor (all optional, non-blocking)

• set timestamp not restored. time_zone is now reset, but timestamp is left overridden. It's the last block in the file so there's no bleed, but for symmetry consider set timestamp=default; alongside the time_zone reset.
• Duplicated my_error block — identical to the date_add_interval overflow handler earlier in the same function; could be factored into a small helper, but fine as-is given it's a fixup.
• Comment nit: with the check scoped to TIMESTAMP, /* Likely Year 2038/2106 Problem */ — "Likely" is no longer needed; the cause is definite.

Performance & security

• No performance impact (one branch on the cold partition-creation path); no security concerns.

Summary

This revision resolves every substantive point raised across earlier revisions — the fix is correct and well-scoped, and the test is now consistent, meaningful (error path + atomicity + DATE/DATETIME success), and deterministic. Ready to squash. The only leftovers are cosmetic: restoring timestamp, the duplicated my_error, and the "Likely" wording — none of which need to block.

[spetrunia]

Ok to push

[mariadb-YuchenPei]

Thanks for the review. Squashed into MDEV-15621/#5152 and will push there.