Improve README with extra info about concurrency (#520)

Author: cedricpim

README.md

@@ -440,7 +440,7 @@ class MyJob < ApplicationJob
 
 When a job includes these controls, we'll ensure that, at most, the number of jobs (indicated as `to`) that yield the same `key` will be performed concurrently, and this guarantee will last for `duration` for each job enqueued. Note that there's no guarantee about _the order of execution_, only about jobs being performed at the same time (overlapping).
 
-The concurrency limits use the concept of semaphores when enqueuing, and work as follows: when a job is enqueued, we check if it specifies concurrency controls. If it does, we check the semaphore for the computed concurrency key. If the semaphore is open, we claim it and we set the job as _ready_. Ready means it can be picked up by workers for execution. When the job finishes executing (be it successfully or unsuccessfully, resulting in a failed execution), we signal the semaphore and try to unblock the next job with the same key, if any. Unblocking the next job doesn't mean running that job right away, but moving it from _blocked_ to _ready_. Since something can happen that prevents the first job from releasing the semaphore and unblocking the next job (for example, someone pulling a plug in the machine where the worker is running), we have the `duration` as a failsafe. Jobs that have been blocked for more than duration are candidates to be released, but only as many of them as the concurrency rules allow, as each one would need to go through the semaphore dance check. This means that the `duration` is not really about the job that's enqueued or being run, it's about the jobs that are blocked waiting.
+The concurrency limits use the concept of semaphores when enqueuing, and work as follows: when a job is enqueued, we check if it specifies concurrency controls. If it does, we check the semaphore for the computed concurrency key. If the semaphore is open, we claim it and we set the job as _ready_. Ready means it can be picked up by workers for execution. When the job finishes executing (be it successfully or unsuccessfully, resulting in a failed execution), we signal the semaphore and try to unblock the next job with the same key, if any. Unblocking the next job doesn't mean running that job right away, but moving it from _blocked_ to _ready_. Since something can happen that prevents the first job from releasing the semaphore and unblocking the next job (for example, someone pulling a plug in the machine where the worker is running), we have the `duration` as a failsafe. Jobs that have been blocked for more than duration are candidates to be released, but only as many of them as the concurrency rules allow, as each one would need to go through the semaphore dance check. This means that the `duration` is not really about the job that's enqueued or being run, it's about the jobs that are blocked waiting. It's important to note that after one or more candidate jobs are unblocked (either because a job finishes or because `duration` expires and a semaphore is released), the `duration` timer for the still blocked jobs is reset. This happens indirectly via the expiration time of the semaphore, which is updated.
 
 
 For example:
@@ -473,7 +473,7 @@ class Bundle::RebundlePostingsJob < ApplicationJob
 
 In this case, if we have a `Box::MovePostingsByContactToDesignatedBoxJob` job enqueued for a contact record with id `123` and another `Bundle::RebundlePostingsJob` job enqueued simultaneously for a bundle record that references contact `123`, only one of them will be allowed to proceed. The other one will stay blocked until the first one finishes (or 15 minutes pass, whatever happens first).
 
-Note that the `duration` setting depends indirectly on the value for `concurrency_maintenance_interval` that you set for your dispatcher(s), as that'd be the frequency with which blocked jobs are checked and unblocked. In general, you should set `duration` in a way that all your jobs would finish well under that duration and think of the concurrency maintenance task as a failsafe in case something goes wrong.
+Note that the `duration` setting depends indirectly on the value for `concurrency_maintenance_interval` that you set for your dispatcher(s), as that'd be the frequency with which blocked jobs are checked and unblocked (at which point, only one job per concurrency key, at most, is unblocked). In general, you should set `duration` in a way that all your jobs would finish well under that duration and think of the concurrency maintenance task as a failsafe in case something goes wrong.
 
 Jobs are unblocked in order of priority but queue order is not taken into account for unblocking jobs. That means that if you have a group of jobs that share a concurrency group but are in different queues, or jobs of the same class that you enqueue in different queues, the queue order you set for a worker is not taken into account when unblocking blocked ones. The reason is that a job that runs unblocks the next one, and the job itself doesn't know about a particular worker's queue order (you could even have different workers with different queue orders), it can only know about priority. Once blocked jobs are unblocked and available for polling, they'll be picked up by a worker following its queue order.