test(state): make the quarantine-guard test exercise the guard
The seed put the retry deadline 900s in the future, so on unguarded code the method short-circuited on the backoff check and the protection assertions passed anyway; only the field-reset assertions failed. Seed the deadline in the past so a rebuild is due, and condense the guard comment.
This commit is contained in:
+6
-18
@@ -682,24 +682,12 @@ class SessionSchemaMixin:
|
||||
if not getattr(self, "_fts_stale", False):
|
||||
return False
|
||||
if getattr(self, "_db_corrupt", False):
|
||||
# Quarantined: structural corruption was already observed on
|
||||
# this handle, so the only safe policy is to stop touching the
|
||||
# file (mirrors hermes_state.py's _try_wal_checkpoint). A full
|
||||
# FTS rebuild is real DDL/DML against the same damaged image —
|
||||
# exactly what quarantine exists to prevent — and this method
|
||||
# is called unconditionally every housekeeping tick for the
|
||||
# life of a long-running gateway process, so a stale-FTS flag
|
||||
# left set on a now-corrupt handle would otherwise retry the
|
||||
# rebuild forever.
|
||||
#
|
||||
# Reset the backoff bookkeeping too (mirrors the success path's
|
||||
# own reset a few lines down): no code path today clears
|
||||
# _db_corrupt on a live handle, so this is inert in practice,
|
||||
# but leaving a doubled _fts_stale_retry_interval sitting behind
|
||||
# a flag nothing currently clears is a footgun for whoever adds
|
||||
# an un-quarantine/recovery path later — the next real retry
|
||||
# should start from the default backoff, not wherever this
|
||||
# handle's interval happened to be when it was quarantined.
|
||||
# Quarantined: never run FTS DDL/DML against a damaged image
|
||||
# (mirrors _try_wal_checkpoint / close). This runs every
|
||||
# housekeeping tick for the life of a gateway process, so a stale
|
||||
# flag on a corrupt handle would otherwise retry the rebuild
|
||||
# forever. Reset the backoff so any future un-quarantine path
|
||||
# starts from the default interval, not a doubled stale one.
|
||||
self._fts_stale_retry_after = 0.0
|
||||
self._fts_stale_retry_interval = 0.0
|
||||
return False
|
||||
|
||||
@@ -648,9 +648,12 @@ class TestDeferredFtsRetryInProcess:
|
||||
assert gw._fts_stale is True
|
||||
gw._db_corrupt = True
|
||||
gw._db_corrupt_reason = "database disk image is malformed"
|
||||
# Simulate a handle that had already been backing off for a
|
||||
# while before it tripped quarantine.
|
||||
gw._fts_stale_retry_after = time.monotonic() + 900.0
|
||||
# A retry that is DUE (backoff already elapsed) on a handle that
|
||||
# had been backing off before it tripped quarantine. Seeding the
|
||||
# deadline in the past matters: a future deadline would make the
|
||||
# unguarded code short-circuit on the backoff check and this test
|
||||
# would pass without the quarantine guard ever being exercised.
|
||||
gw._fts_stale_retry_after = time.monotonic() - 1.0
|
||||
gw._fts_stale_retry_interval = 900.0
|
||||
assert gw.retry_deferred_fts_recovery() is False
|
||||
# Untouched: still marked stale, triggers still absent — no
|
||||
|
||||
Reference in New Issue
Block a user