Dependencies Among Jobs¶
When editing a job, it is possible to specify the job(s) this one depends on. The cron time trigger has a section for specifying this purpose. Each trigger may have its own dependencies. By default, a trigger will not execute the job unless the jobs it depends on have successfully finished their execution (including all their retries, if configured). This behavior may be altered by checking the Execute this job even though any of its dependencies finish with error check box, in order to let the job be executed having its dependencies finished successfully or not. Optionally, the maximum time (in milliseconds) for waiting for the dependencies to be completed can be specified.
It is important to note that jobs cannot depend on jobs that are on draft state. Nevertheless, draft jobs may have dependencies (and they will be kept if the project is exported).
If a job is waiting for a trigger’s dependencies to be satisfied and then the jobs is executed as consequence of other trigger, then the wait for the first trigger dependencies is reset.
Example: Suppose a job “A”, having a trigger that depends on a job “B”, starts its execution. Then, “A” will be in WAITING state until job “B” completes its execution (note that if “B” is already being executed when “A” starts, then “A” will be in WAITING state until “B” starts a new execution and finishes it). If job “B” has retries (because of the Retry Handler configuration), “A” will keep waiting until all the retries of “B” are executed. When “B” finishes,
- If it finishes successfully, then job “A” will be executed.
- If it finishes with error, then job “A” will stop waiting and will not be executed (nor the jobs that follows “A” in the chain of dependencies). A message informing about this circumstance will be added to the report.
- If it finishes with error and the job “A” was configured to be executed even though any of its dependencies finish with error, then job “A” will be executed.
If “A” and “B” are configured to be executed at the same time, sometimes “B” could start before “A”, and other times “A” could start before “B”:
- If “A” starts before “B”, then “A” is executed when “B” completes its execution
- If “B” starts before “A”, then “A” is executed when “B” completes its next execution (not the current one). This means that “A” will be in WAITING state until the next time job “B” is scheduled to run.
This way, if “A” depends on “B” and “B” depends on “C”, then “A” should be scheduled to be the first job to be executed. The following, would be an example of scheduling:
- “A” starts at time X.
- “B” starts at time X + 1 second.
- “C” starts at time X + 2 seconds.
If the timeout of A is reached before B finishes, the behavior will be like in b), with the message “Timeout exceeded for this job, while waiting for dependencies to be satisfied.” added to the report.
If while B is being executed, another trigger causes the execution of A, then the first trigger waiting is reset. Therefore, when B finishes, A will not be executed. Another execution of B starting and ending while A is waiting would be required for the dependencies of the first trigger to be satisfied.
If “B” were deleted, the configuration of job “A” would be updated (removing this dependency from the trigger/s that have it) and would change to the DISABLED state (although if “A” were in the RUNNING state, it would continue execution until finishing and, then, would change to DISABLED state). The user must then enable or edit the job to re-schedule it.