Repository navigation
Ufid handle - #6367
Ufid handle#6367akshatsikarwar wants to merge 2 commits into
Conversation
roborivers
left a comment
roborivers
left a comment
There was a problem hiding this comment.
Cbuild submission: Success ✓.
Regression testing: Success ✓.
The first 10 failing tests are:
incoherent_slow **quarantined**
reco-ddlk-sql **quarantined**
consumer_non_atomic_default_consumer_generated **quarantined**
1e81abe to
344f0c5
Compare
roborivers
left a comment
roborivers
left a comment
There was a problem hiding this comment.
Cbuild submission: Success ✓.
Regression testing: Success ✓.
The first 10 failing tests are:
constraints_partial_index_off_generated [failed with core dumped]
constraints [failed with core dumped]
sqlqueries
tunables
sc_truncate_lockorder_generated [timeout] **quarantined**
reco-ddlk-sql [timeout] **quarantined**
skipscan [timeout] **quarantined**
roborivers
left a comment
roborivers
left a comment
There was a problem hiding this comment.
Cbuild submission: Success ✓.
Regression testing: 0/758 tests failed ⚠.
The first 10 failing tests are:
reco-ddlk-sql [timeout] **quarantined**
The ufid hash holds one handle per file. Recovery and replication can open several DB_AM_RECOVER handles for the same file: __ufid_open handles taken while applying ufid-logged records, handles dbreg opens during recovery (e.g. after verify-match on a master swing), and handles registered when a replicant becomes master. When a newer handle is registered, __ufid_add_dbp drops the older one from the hash and nothing closes it. Two leaks follow: - __db_clear_ufid_hash (TRUNCATE, or an alter that rebuilds files) closes only the handle currently in the hash. Dropped handles keep the retired btree files open, so delfiles fails or the unlinked files keep using disk space until the process restarts. - __ufid_open handles have no owner other than the hash, so a clean exit fails with "Database handles open during environment close". Every DB_AM_RECOVER handle goes through __ufid_add_dbp when it opens and stays on dbenv->dblist until it closes. So instead of tracking dropped handles, __db_close_recover_dbps walks the dblist and closes every recovery handle for a fileid, or for all files. __db_clear_ufid_hash uses it for the retired file, and __dbenv_close uses it for all files after dbreg has closed its own handles. Add ufid_handle_leak and ufid_deleted_file_leak cluster tests. Signed-off-by: Akshat Sikarwar <asikarwar1@bloomberg.net>
ufid_recovery, ufid_reopen and dbreg_race ran /opt/bb/bin/cdb2sql over ssh, and catchup restarted a node with /opt/bb/bin/comdb2, so they failed on clusters without an installed comdb2. Use CDB2SQL_EXE and COMDB2_EXE, which setup already relies on being present at the same path on every node. Signed-off-by: Akshat Sikarwar <asikarwar1@bloomberg.net>
344f0c5 to
21dd2c2
Compare
roborivers
left a comment
roborivers
left a comment
There was a problem hiding this comment.
Cbuild submission: Success ✓.
Regression testing: Success ✓.
The first 10 failing tests are:
nowritetimeout
No description provided.