Fix dumping on exit

[mikiw]

Usage related changes

• Quick fix for double execution of dump function.

Development related changes

• test_dumping_on_exit updated to cover scenario with call after load

Checklist:

• Applied formatting - ./scripts/format.sh
• No linter errors - ./scripts/lint.sh
• Performed code self-review
• Rebased to the last commit of the target branch (or merged it into my branch)
• Documented the changes
• Linked the issues which this PR resolves
• Updated the tests
• All tests are passing

[mikiw]

The second dump was performed with a clean instance of starknet which means that this functionality didn't work at all.

This is related to the introduction of GunicornServer and the number of workers:

    asyncio.run(state.starknet_wrapper.initialize())

    try:
        print(f" * Listening on https://{args.host}:{args.port}/ (Press CTRL+C to quit)")
        pid = os.getpid()
        GunicornServer(app, args).run()
    except KeyboardInterrupt:
        pass
    finally:
        if args.dump_on == DumpOn.EXIT and os.getpid() != pid:
            state.dumper.dump()
            sys.exit(0)

Although a new test was introduced and a quick fix was made it seems that it's a hack, not a real solution.

After starting a starknet devnet there are 2 processes an instance of GunicornServer and 1 worker. The GunicornServer instance has an empty devnet and the worker has a real instance with blocks.

The number of processes is equal to the count of workers + 1.

For example, if we increase the number of workers from 1 to 2 test will fail.

[mikiw]

@Solpatium do you have any idea how to fix it properly?

[FabijanC]

Probably due to this line each process has its own "global" state
https://github.com/Shard-Labs/starknet-devnet/blob/40d5b6f75fd80e5adea989e5f45ebb76c5c25cde/starknet_devnet/state.py#L57

[mikiw]

Should we use worker_exit in gunicorn and always use 1 worker?

Add to config:

self.cfg.set("worker_exit", worker_exit)

Example:

def worker_exit(server, worker):
    try:
        print("worker_exit", os.getpid())
        state.dumper.dump()
        sys.exit(0)
    except:
        pass

[Solpatium]

Having more than 1 worker doesn't make sense at the moment as there is no way of sharing the state between them, right?

[FabijanC]

Having more than 1 worker doesn't make sense at the moment as there is no way of sharing the state between them, right?

Yes. If you don't have objections, I'll merge the PR as it currently is.