On Wed, Dec 02, 2020 at 02:44:31PM -0500, Stephen Frost wrote:
> And committed.
This has been committed as of dc11f31a, that changed the configuration
of the node in the kerberos test to use logging_collector. Wouldn't
it be simpler to not use the logging collector here and use a logic
similar to what we do in PostgresNode::issues_sql_like() where we
truncate the log file before checking for a pattern?
It seems to me that this would make the tests faster, that the test
would not need to wait for the logging collector and that the code
could just use slurp_file($node->logfile) to get the data it wants to
check for a given pattern without looking at current_logfiles. I also
think that not using truncate() on the logfile generated has the
disadvantage to make the code fuzzy for its verification once we
introduce patterns close to each other, as there could easily be an
overlap. That's one problem that SQL pattern checks had to deal with
in the past. Thoughts?
--
Michael