Page MenuHomeSoftware Heritage

Add a keyspace_queue fixture to prepare keyspaces concurrently.
AbandonedPublic

Authored by vlorentz on Fri, Jan 31, 4:06 PM.

Details

Reviewers
ardumont
Group Reviewers
Reviewers
Summary

Keyspace creation + initialization with a schema takes about 60% of the
total time to run a test.
This new fixture runs keyspace creation concurrently to tests,
so other tests can run while a keyspace is being created.

Depends on D2566.

Diff Detail

Repository
rDSTO Storage manager
Branch
keyspace_queue
Lint
No Linters Available
Unit
No Unit Test Coverage
Build Status
Buildable 10412
Build 15499: tox-on-jenkinsJenkins
Build 15498: arc lint + arc unit

Event Timeline

vlorentz created this revision.Fri, Jan 31, 4:06 PM
ardumont accepted this revision.Mon, Feb 3, 10:29 AM
ardumont added a subscriber: ardumont.

Worth a try i guess.

The diff's job seems to fail though (triggered a rebuild from jenkins).

Trying to check the difference on my box as well.

This revision is now accepted and ready to land.Mon, Feb 3, 10:29 AM

Build has FAILED

i reproduce this here as well...

???
E cassandra.cluster.NoHostAvailable: ('Unable to connect to any servers', {'127.0.0.1:59493': OperationTimedOut('errors=None, last_host=None')})

Seems to be the root of the problem.
Initial tests seem to be fine up to the point where the connection refuses to establish again.

Also, that's not the only problem i had.
On another run which got further, hypothesis found a falsifiable sample...
Relatedly to timing apparently.

Falsifying example: test_add_arbitrary(
    self=<swh.storage.tests.test_cassandra.TestCassandraStorageGeneratedData at 0x7fed490336d8>,
    swh_storage=<swh.storage.cassandra.storage.CassandraStorage at 0x7fed0c2c47b8>,
    objects=[('content',
      Content(sha1=b'\t&\xc1\xd5!f\xd1O\x1879\xb3\xe0\x94\xeb\xcb\xf9d\x1bl', sha1_git=b'\xf6\x93\xe5\x00\xe2\xa4\xc4\xc9<\x12\x1d\x8d\x0eN\xb9\x01O\xca\x11<', sha256=b'\xaf\xbd\xc7\xf4Nt8@u\nQ\xcbN\x96|Q\xd1}\xf9]\xa8\xbf\xb6\x11"\xabHV\x16[\xa2\x85', blake2s256=b'\t\x07\xc2\xec\x8c \xd9`\x11\xc0\xe6\xe3\xf8\xa2\xd6\x95=\xf3\xccE\xff\xfa:\x1d\xbf\x08\xee\xf3q\x0b\xe4J', length=60, status='absent', reason='&\x1b', data=None, ctime=None)),
     ('origin_visit',
      OriginVisit(origin='http://m0w-.o0k-.iq59897v', date=datetime.datetime(593, 3, 28, 1, 1, 58, 553680), status='partial', type='\x0c"\x06\x08\x18!\U00062918\x1f\x19\U0009e77f', snapshot=b'', metadata=None, visit=351))],
)
Unreliable test timings! On an initial run, this test took 6727.64ms, which exceeded the deadline of 5000.00ms, but on a subsequent run it took 11.29 ms, which did not. If you expect this sort of variability in your test timings, consider turning deadlines off for this test by setting deadline=None.

It may have fallen on the case in-between keyspace creation so that took more time?

It just means Cassandra crashed in the middle of the tests.

vlorentz updated this revision to Diff 9350.Mon, Feb 3, 11:56 AM
  • Enable logging

This is stupid. Replaced by D2615.

vlorentz abandoned this revision.Mon, Feb 3, 12:26 PM