GitLab

NOTE: This should be backported to the previous branches, but the latter are missing increment_a_counter()

Check that the analysis is not BLOCKED before running a special-batch (i.e. runWorker with a job_id)

This is can be overriden with "-force 1"

experimental: detect and bury the UNKWN workers -- the user takes responsibility by running -unkwn flag

bugfix: switching to using the newly created find_all_by_pattern() and collections in Worker to fix the -job_id bug

bugfix to [to bee183d1] : ignore a possible collision with another beekeeper updating worker_resource_usage

schema_change: removed analysis_stats[_monitor].num_required_workers, substituted with AS::estimate_num_required_workers()

bugfix: stop registering some NO_WORK workers as UNKNOWN, by changing the order of Valley/DB reconciliation

The problem was when a worker is contaminated and has to die: remainings jobs
of the batch were not released.
We now have an explicit list of all the cause_of_death that could happen through
a batch, and for which we should free the jobs

Added 'meadow_user' to Worker object+schema which is now routinely recorded. Also renamed 'host' to 'meadow_host' for consistency

"self-burial" had two meanings:
 - should we release the undone jobs ?
 - should we update the last_check_in field in the worker table ?
The former can simply be found by checking whether the worker already has a
role attached to it.
The latter is only set when a worker registers its own death, and is not set
by the GarbageCollector.

"self-burial" had two meanings:
 - should we release the undone jobs ?
 - should we update the last_check_in field in the worker table ?
The former can simply be found by checking whether the worker already has a
role attached to it.
The latter is only set when a worker registers its own death, and is not set
by the GarbageCollector.

(and probably RUNLIMIT as well)
The problem was in Queen::register_worker_death(). The code fetches the
current role, finalizes it, and then updates the worker table. The role
that is fetched from the database with fetch_last_unfinished_by_worker_id()
is not connected to any workers and later, when
AnalysisJobAdaptor::release_undone_jobs_from_role() needs to know the cause
of death and the number of work done by the worker, the role would load
another instance of the worker using the information currently stored in
the database, that is: before it has been updated.
Hence, the code was complaining that cause_of_death was not populated.
By associating the role to the worker, we allow the method to read the
existing information

API change: schedule_workers() now also returns the "plan" used by suggest_analysis_to_specialize_a_worker(); only one return format for schedule_workers().