PVF: Add back socket path parameter, use tmp socket path

[mrcnski]

Problem

Running zombienet locally:

WARN tokio-runtime-worker parachain::pvf: cannot bind unix socket: Custom { kind: InvalidInput, 
error: "path must be shorter than libc::sockaddr_un.sun_path" } 
debug_id=prepare program_path="/home/alexggh/repos/polkadot-sdk/target/testnet/polkadot-prepare-worker" 
extra_args=["prepare-worker", "--node-impl-version", "1.1.0"] 
worker_dir=WorkerDir { path: "/tmp/zombie-a609ef68a12d8517c8174b62ee3a0899_-484651-6nFd48QbCHfO/alice/data/chains/rococo_local_testnet/db/full/pvf-artifacts/worker-dir-prepare-2Q4B1v7qio" } 
socket_path="/tmp/zombie-a609ef68a12d8517c8174b62ee3a0899_-484651-6nFd48QbCHfO/alice/data/chains/rococo_local_testnet/db/full/pvf-artifacts/worker-dir-prepare-2Q4B1v7qio/socket"

This is because by default zombienet decides to prefix things with /tmp/zombie-a609ef68a12d8517c8174b62ee3a0899_-484651-6nFd48QbCHfO and that gets us beyond the socket limitation size.

Fix

Looks like the socket path limit is 108 bytes. This should definitely be fixed, users are allowed to use an arbitrary cache path. There is no need for the socket to be under the cache path though, we can revert to using a tmp location like we were before.

Related

Closes #1765
This was introduced with #1373

[alindima]

LGTM, minus the failing test

[s0me0ne-unkn0wn]

@pepoviola what's the rationale behind having that long directory suffixes in the first place?

[s0me0ne-unkn0wn]

@mrcnski strictly speaking, the proposed fix fixes the case, but not the entire problem. That weird sun_path limit is the reason why it is a good practice to put socket files under /run, /var/run, or directly under /tmp. Is there any reason why shouldn't we follow that practice?

Ah yes, we're running as non-root, and /run is not guaranteed to be writeable. But why not just /tmp then?

[s0me0ne-unkn0wn]

LGTM as a quick fix, but probably better approach should be considered in the future.

[pepoviola]

@pepoviola what's the rationale behind having that long directory suffixes in the first place?

The idea was to have some pattern like namespace_random but I think we can shrink to only use the namespace that will be already pretty unique. I will open an issue to track this.
Thanks!

[mrcnski]

@mrcnski strictly speaking, the proposed fix fixes the case, but not the entire problem. That weird sun_path limit is the reason why it is a good practice to put socket files under /run, /var/run, or directly under /tmp. Is there any reason why shouldn't we follow that practice?

Ah yes, we're running as non-root, and /run is not guaranteed to be writeable. But why not just /tmp then?

We're using std::env::temp_dir (in with_transient_socket_path we call tmppath which calls temp_dir), which should return /tmp:

On Unix, returns the value of the TMPDIR environment variable if it is set, otherwise for non-Android it returns /tmp.