You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Flake: oracle corpus case fails once with "I/O error: No such file or directory" applying a rule in the subscription group engine #401
Seen once on 2026-10-06 in a local make unit-test run (devbox, lean debug build, branch fm/t-bundle-b on top of 2ed4a4a; the oracle changes on that branch add a corpus pack and an optional adapter, not this path):
iql_corpus_agrees_with_spec_and_reference panicked at tests/differential_oracle/corpus.rs:333:5:
1 divergences:
examples/iql/14_aggregations/16_empty_agg_edge_cases.iql.out: Outcome divergence in 'subscription[group]' vs 'recompute' at `+empty_avg(avg<V>) <- filtered(_, V)` (revision 5): Rejected("I/O error: No such file or directory (os error 2)"), baseline Applied
The subscription[group] adapter's engine (its own Handler on a private /dev/shm data directory, tests/differential_oracle/engine.rs) refused a persistent-rule insert with ENOENT, while the recompute engine applied it. Another oracle run (INPUTLAYER_ORACLE_VIEWS=maintained, which panics on startup) was running at the same time on the host, but no oracle path is shared between processes. Four reruns of the oracle and a full rerun of the suite passed.
An ENOENT on a rule commit points at a persistence write whose file or directory went missing under it (a temp file renamed twice, or a directory created lazily by another thread). Per the flake policy in TESTING.md: find the write that can see ENOENT, or mark the case Reproduction::Racy with this issue until it is fixed.
Likely the same cause as #402. The oracle binary runs several engines in one process. A rule commit calls revisions::reserve_ahead, which fails with ENOENT when another open engine in the process cannot write its revision bound, for example one whose temporary data directory was removed before the engine dropped. The refused engine's own files are not involved.
Activity
jsam commentedon Oct 6, 2026
Likely the same cause as #402. The oracle binary runs several engines in one process. A rule commit calls
revisions::reserve_ahead, which fails with ENOENT when another open engine in the process cannot write its revision bound, for example one whose temporary data directory was removed before the engine dropped. The refused engine's own files are not involved.