eunit tests don't cope with the lack of O_SYNC support in OTP 16. Wa

<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="25

eunit tests fail in OTP16 about leveled HOT 3 CLOSED

martinsumner commented on August 20, 2024

eunit tests fail in OTP16

from leveled.

Comments (3)

martinsumner commented on August 20, 2024

Having resolved the sync issue, the most noticable change was that tests began timing out.

Running tests in 18.3:

Top 10 slowest tests (11.943 seconds, 70.3% of total time):
  leveled_iclerk:file_gc_test/0
    2.210 seconds
  leveled_inker:compact_journal_test/0
    2.047 seconds
  leveled_pmanifest:snapshot_timeout_test/0
    2.005 seconds
  leveled_sst:simple_persisted_test/0
    1.212 seconds
  leveled_tree:skpl_timing_test/0
    1.130 seconds
  leveled_inker:empty_manifest_test/0
    1.011 seconds
  leveled_pmem:with_index_test_/0
    0.747 seconds
  leveled_pclerk:merge_file_test/0
    0.579 seconds
  leveled_cdb:state_test/0
    0.514 seconds
  leveled_tictac:merge_bysize_xlarge_test_/0
    0.488 seconds

Finished in 16.985 seconds
132 tests, 0 failures

But running tests in R16 (basho10):

Top 10 slowest tests (72.099 seconds, 85.3% of total time):
  leveled_bookie:hashlist_query_test_/0
    13.720 seconds
  leveled_iclerk:compact_singlefile_totwosmallfiles_test_/0
    11.224 seconds
  leveled_cdb:get_keys_byposition_manykeys_test_/0
    11.184 seconds
  leveled_bookie:foldobjects_vs_foldheads_bybucket_test_/0
    8.958 seconds
  leveled_bookie:foldobjects_vs_hashtree_test_/0
    8.911 seconds
  leveled_bookie:hashlist_query_withjournalcheck_test_/0
    8.892 seconds
  leveled_inker:compact_journal_test/0
    2.685 seconds
  leveled_bookie:ttl_test/0
    2.305 seconds
  leveled_iclerk:file_gc_test/0
    2.214 seconds
  leveled_pmanifest:snapshot_timeout_test/0
    2.006 seconds

Finished in 84.511 seconds
132 tests, 0 failures

from leveled.

martinsumner commented on August 20, 2024

This time difference may be related to the riak_sync of flushing manually each write (rather than passing the sync flag). This might be macbook specific.

Essentially running sync on OT18 (which should pass the sync flag) is only marginally slower (20%) than running the ct tests with no sync on OTP16. It is definitely syncing - as it manages >> 1000 tps in iostat. The same test with sync (and hence riak_sync) in OTP 16 only manages 500 tps and is more than fifty times slower

Without sync enabled there is negligible performance difference between OTP versions.

from leveled.

martinsumner commented on August 20, 2024

#89

from leveled.

Related Issues (20)

Recommend Projects