In most cases, an IPC mechanism should be better. You open up named pipes (which can be anything, so why not JSONL) for two way communication, and use that. Or shared memory of some other kind. Only one of the processes would actually write the file to disk once in a while, to avoid on-disk corruption and allow the saving program to do checks and corrections.

But if really needed, here are some thoughts: Theoretically, you could write JSONL (JSON Lines), because that’s simple and human-readable in cases of debugging. If process A wants to write a new line, a new event, for example, it has to check for “'\r\n” at the end, and write-append it in one go. Process B gets the filesystem event that the file changed and reads it. It would be better of course, to create a simple 0-bytes “.lock” file while the writing is going on and removed it afterwards. To avoid corruption, you could also create checksums and whatever. Depending on your use case, that can work. I mean, I open one file sometimes in two editors and you can loose of course modifications that you did while it reloads the file from the other editor.