search

scylladb / cql-stress

8 stars 4 forks source link

main: set max_retries_per_op to 9 (c-s default) #64

Closed muzarski closed 6 months ago

muzarski commented 6 months ago

Motivation

By default, cassandra-stress takes 9 attempts to perform an operation before ending the benchmark with failure.

Before this PR, cql-stress would fail immediately after first failed operation. This resulted in SCT run failures (SCT + cql-stress) when some nemesis would kill one of the nodes in the cluster:

   7689156,   32227,     0.9,     0.8,     1.6,     2.1,     4.7,     9.0,  256.0,      0
   7720832,   31675,     0.9,     0.9,     1.6,     2.1,     3.6,     7.3,  257.0,      0
   7753593,   32738,     0.9,     0.8,     1.5,     2.0,     3.9,     7.3,  258.0,      0
2023-12-15T15:22:44.866516Z ERROR cql_stress_cassandra_stress::operation::read: read error error=Invalid message: Frame error: early eof partition_key=Blob([80, 48, 51, 53, 55, 54, 75, 75, 49, 49])
2023-12-15T15:22:44.866544Z ERROR cql_stress_cassandra_stress::operation::read: read error error=Invalid message: Frame error: early eof partition_key=Blob([78, 57, 55, 55, 76, 56, 48, 75, 54, 48])
Error: An error occurred during the benchmark

Caused by:
    Invalid message: Frame error: early eof

When changing the max_retries_per_op to c-s default, the SCT run finishes successfully.

Changes