GitLab

added a new protected_prepare_execute() method to avoid deadlocks and used it twice in AnalysisJobAdaptor, to fix Stephen's deadlocks

bugfix: if re-running a job that creates a semaphored group, we no longer die (thanks Miguel for reporting)

feature: jobless workers will now leave module compilation errors in the job_message table (thanks, Kathryn!)

merge reset_and_grab into one subroutine; pre-increment dependent semaphore if re-running a DONE job; use -force flag for force-running an individual job

moved special-job-reset and special-job-reclaim into the same call, removed the unnecessary fetch in between

extend the param_init() of the garbage-collected jobs to include analysis->parameters() for template substitution (still limited!)

introduced semaphored_job_count, renamed unclaimed_job_count-->ready_job_count, changed reporting, fixed hive_capacity=0

moved failed_job_tolerance, max_retry_count, can_be_empty and priority columns from analysis_stats to analysis_base

bugfix: changed the order of the atomic SEMAPHORED->READY state&counter UPDATE so that it works as intended in SQLite as well

Dropped 'BLOCKED' job status and introduced 'SEMAPHORED' status that is maintained in sync with semaphore_counts; less confusing and more efficient (with new 3-part index)

if running a worker with a specific job_id, the status is set to READY, but the retry_count is set depending on whether PRE_CLEANUP is needed or not

At last rename GET_INPUT into FETCH_INPUT for consistency between the schema and the code (it seems to be harder to patch all the accumulated code)

simplified logic that controls how semaphores are propagates (preparing for semaphore escaping rule support)