FEATURE: Partitioned Wait Lists

[skurfuerst]

If you have a multi-tenant application, you might want to use one wait-list per tenant (e.g. for import jobs),
combined with global concurrency limits (depending on the globally available server resources).

To enable this, do the following:

• queue_strategy: partitioned_replace: Enabled partitioned wait list
• queue_partition_limit: 1 (or higher): Configure the number of wait-list slots per tenant. The last slot gets replaced
  when the wait-list is full.
• can be combined with arbitrary start_delay and concurrency as expected

Full example:

pipelines:
  do_something:
    queue_strategy: partitioned_replace
    # prevent starvation
    queue_partition_limit: 2
    concurrency: 1
    # Queues a run of the job and only starts it after 10 seconds have passed (if no other run was triggered which replaced the queued job)
    start_delay: 10s
    tasks: # as usual

Additionally, when submitting a job, you need to specify the queuePartition argument:

POST /pipelines/schedule

{
  "pipeline": "my_pipeline",
  "variables": {
    ...
  },
  "queuePartition": "tenant_foo"
}

[skurfuerst]

also contains lots of cleanups and a critical bugfix (87d979e)