Uploaded image for project: 'WiredTiger'
  1. WiredTiger
  2. WT-3995

Enhance timestamp abort to accept more number of threads

    • Type: Icon: Bug Bug
    • Resolution: Fixed
    • Priority: Icon: Major - P3 Major - P3
    • 4.0.4, 4.1.4, WT3.2.0
    • Affects Version/s: None
    • Component/s: None
    • Labels:
    • 2
    • Storage Engines 2018-09-24, Storage Engines 2018-10-08

      The timestamp abort test allows a maximum of 12 threads. We should change it so more threads can be configured. The naive change to do that leads to some test failures when high thread counts are chosen, due to session max being exceeded:

      ...
      test_timestamp_abort: FAILED: thread_run/282: td->conn->open_session(td->conn, NULL, NULL, &session): WT_ERROR: non-specific WiredTiger error[1522032138:76578][12961:0x7f4753fff700], WT_CONNECTION.open_session: out of sessions, configured for 134 (including internal sessions): WT_ERROR: non-specific WiredTiger error
      test_timestamp_abort: FAILED: thread_run/293: td->conn->open_session( td->conn, NULL, NULL, &oplog_session): WT_ERROR: non-specific WiredTiger error
      process aborting
      [1522032138:76770][12961:0x7f4627fff700], WT_CONNECTION.open_session: out of sessions, configured for 134 (including internal sessions): WT_ERROR: non-specific WiredTiger error
      test_timestamp_abort: FAILED: thread_run/293: td->conn->open_session( td->conn, NULL, NULL, &oplog_session): WT_ERROR: non-specific WiredTiger error
      process aborting
      
      process aborting
      Parent: compatibility: false, in-mem log sync: false, timestamp in use: true
      Parent: Create 181 threads; sleep 11 seconds
      CONFIG: test_timestamp_abort -h WT_TEST.timestamp-abort -T 181 -t 11
      test_timestamp_abort: FAILED: Child process 12961 abnormally exited: Invalid argument
      

      We should enhance the code to allow more threads and to error reasonably if a number greater than max is manually configured.

      It would also be interesting to facilitate creating workloads that do a lot of page splits, since they tend to uncover data consistency problems. I've made such a change locally before by making the following code change:

      --- a/test/csuite/timestamp_abort/main.c
      +++ b/test/csuite/timestamp_abort/main.c
      @@ -425,12 +425,13 @@ run_workload(uint32_t nth)
              /*
               * Create all the tables.
               */
      +#define TABLE_FORMAT_SHARED    "key_format=S,value_format=u,internal_page_max=1k,leaf_page_max=1k,memory_page_max=64k,allocation_size=512,"
              testutil_check(session->create(session, uri_collection,
      -               "key_format=S,value_format=u,log=(enabled=false)"));
      +               TABLE_FORMAT_SHARED "log=(enabled=false)"));
              testutil_check(session->create(session,
      -           uri_local, "key_format=S,value_format=u"));
      +           uri_local, TABLE_FORMAT_SHARED));
              testutil_check(session->create(session,
      -           uri_oplog, "key_format=S,value_format=u"));
      +           uri_oplog, TABLE_FORMAT_SHARED));
              /*
               * Don't log the stable timestamp table so that we know what timestamp
               * was stored at the checkpoint.
      
      

      We could do something similar behind a configuration option, or allow the format to be specified on the command line.

            Assignee:
            neha.khatri Neha Khatri
            Reporter:
            alexander.gorrod@mongodb.com Alexander Gorrod
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated:
              Resolved: