watch-task-protocol

watch-task-protocol

The main goal of the watch-task-protocol is making it possible to coordinate multiple watch processes. These are processes that don't exit after a build, but instead continue to rebuild as soon as file-changes are detected. The problem with such processes is that they cannot coordinate themselves with each other. So if both lib and app detect filechanges at the same time, both processes will immediately start re-building. This is a problem, because re-building app only makes sense after lib is built successfully.

So what I came up with to solve this problem is a protocol for such processes to communicate with the task coordinator or task runner. With this protocol, a watch process can notify the runner about changes, and the runner can notify the process when it's time to re-start the task. Additionally, the task can tell the runner if it was successful or not.

The cool thing is that processes not implementing the protocol are still fully compatible with the protocol. Not implementing the protocol means that a process exits after the task is complete and that it's restarted to run the task again. Only if a process wants to stay alive and still tell the task runner that its job is completed, it needs make use of the watch-task-protocol.

The protocol is very simple:

1. When the process starts, immediately runs the task
2. After the task is successful, write WATCH_TASK_PROTOCOL:SUCCEEDED to STDOUT, or if the task failed write WATCH_TASK_PROTOCOL:FAILED
3. Do not restart the task until you receive WATCH_TASK_PROTOCOL:START on STDIN, then immediately restart it and continue to 2.
4. A process can exit at any point, this also means the task is completed and the process shall be restarted the next time the task needs to run.
5. Sometimes a watch process knows best, when it's time to re-run the task because it knows its task the best after all. In this case, it can at any point write WATCH_TASK_PROTOCOL:DETECTED_CHANGES to STDOUT. This tells the task runner that it wants to re-run. But remember, it's the task runners job to actually trigger the re-run. So don't restart the task until you receive the WATCH_TASK_PROTOCOL:START message.