GitLab

turning the max_limbo_seconds into a JSON config parameter that can be as Meadow-specific as you wish

safe_drop_database is a new Utils::Test method to wait for Workers to exit, disconnect and drop the DB

bugfix: make sure we only ever attempt cleanup_left_temp_directory() for workers that got past SUBMITTED stage

Otherwise the worker_temp_directory is not expected to have been created.
Also, meadow_host may not be defined yet (but this is a secondary reason).

Workers dying in SUBMITTED state are detected on in dumb Meadows using a built-in max_limbo_seconds timeout

At this stage psql is our custom version, but PERL5LIB is not set to include PgCommon.

For sqlite databases, only show the filename since the directory name is equivalent to the host name in other drivers, which is not displayed

sqlite URLs that point to directories should be assigned a dbname (i.e. a filename) like other drivers

* version/2.4:
  Use File::Glob to get the list of filenames (#21)

* version/2.3:
  Use File::Glob to get the list of filenames (#21)

* Use File::Glob to get the list of filenames

Previously backtick ls command was used to fetch the list of sql patches filenames.
By using backtick it is required to fork a new process. This would potentially be
a problem for a running process that uses large amount of memory. If there is not
enough spare memory left in the system, the fork operation could be aborted because
of "cannot allocate memory" error.

Fixes ENSEMBL-4823.

[ The PR was added to 2.2 but somehow discarded when 2.2 was merged into 2.3 ]

Conflicts:
	modules/Bio/EnsEMBL/Hive/DBSQL/SqlSchemaAdaptor.pm

If the system is very busy, it may take some time to exec a new process, so the worker process may be called beekeeper.pl for a while

Manually-submitted workers are likely to be missing the EHIVE_SUBMISSION_NAME environment variable. Therefore, we should filter out the workers that have this variable set to *another* pipeline

* version/2.4:
  Show the value that is actually going to be stored, not the requested one

introduced -worker_crash_on_startup_prob option for both runWorker.pl and beekeeper.pl (which passes it through to all workers)

introduced -worker_delay_startup_seconds option for both runWorker.pl and beekeeper.pl (which passes it through to all workers)

changed the order of worker status reconciliation: now starting with the DB and check against Meadows