Skip to content

Temporarily disable problematic unmaterialized source test#11224

Merged
guswynn merged 1 commit into
MaterializeInc:mainfrom
guswynn:disabletest15
Mar 14, 2022
Merged

Temporarily disable problematic unmaterialized source test#11224
guswynn merged 1 commit into
MaterializeInc:mainfrom
guswynn:disabletest15

Conversation

@guswynn
Copy link
Copy Markdown
Contributor

@guswynn guswynn commented Mar 14, 2022

To clean up nightly tests and tests on main, we are disabling this test till we have a better solution (#11192, or even more than that)

partially unblocks https://github.com/MaterializeInc/database-issues/issues/3234

Motivation

See above

Testing

  • This PR has adequate test coverage / QA involvement has been duly considered.

@guswynn guswynn requested review from benesch and philip-stoev March 14, 2022 20:11
@guswynn guswynn enabled auto-merge (rebase) March 14, 2022 20:19
@guswynn guswynn merged commit e6c4774 into MaterializeInc:main Mar 14, 2022
@guswynn guswynn deleted the disabletest15 branch March 16, 2022 21:43
DAlperin added a commit to DAlperin/materialize that referenced this pull request May 15, 2026
…lizeInc#11200 + MaterializeInc#11224)

Adds two new parallel_driver_ shapes that collectively cover both
peek-sequencing variants of the read-hold downgrade bug:

  parallel_driver_explicit_txn_no_since_violation
    * Property: peek-no-since-violation
    * Each invocation: 1 autocommit write into its own seed-private
      antithesis_txn_table, then a fresh connection running
      BEGIN -> 8 SELECTs alternating table/MV -> COMMIT.
    * Targets the new peek sequencing (frontend_peek.rs) via explicit
      multi-statement transactions — the in_immediate_multi_stmt_txn=true
      code path the MaterializeInc#11224 fix (MaterializeInc#36403) addressed.
    * Low per-MV pressure, broad invocation-level concurrency.

  parallel_driver_pw_hot_objects
    * Properties: peek-no-since-violation (same as above) +
      prepared-execute-no-since-violation (new — old peek sequencing)
    * Every invocation hammers the SHARED pw_hot_table + pw_hot_mv_count
      + pw_hot_mv_sum + pw_hot_idx_count set for 10s with a 6-action
      weighted mix (INSERT 30, PREPARE+EXECUTE 40, explicit-txn 15,
      baseline SELECT 10, DELETE-old 5).
    * Targets the old peek sequencing (MaterializeInc#11200) via PREPARE+EXECUTE on
      an indexed MV under continuous frontier advance from concurrent
      writes. Existing parallel-workload's exe_prepared runs against
      fragmented per-seed objects (~1/4 of CI nightly's per-MV
      pressure) — the hot driver concentrates all invocations' writes
      on one shared MV so the bug window has a real chance to open.
    * Storage growth bounded by time-based DELETE
      (ts < now() - INTERVAL '5 minutes').

Why two drivers, not one: peek-no-since-violation and
prepared-execute-no-since-violation share the error matcher but are
distinct properties — old and new peek sequencing are different
code paths that happen to share a bug shape, and PR MaterializeInc#36403 only
fixed one of them. Keeping the assertion messages distinct lets
Antithesis triage tell us which path regressed.

Also bundles helper_pw_hot.py with the shared object-name constants
and the SINCE_VIOLATION_PATTERNS / TRANSIENT_PATTERNS classifiers
that both drivers consume.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants