Follow-up on "Support multiple workers for NODEFS /wordpress mounts"

[brandonpayton]

We are in the midst of merging a number of large PRs:

• Support multiple workers for NODEFS /wordpress mounts #2231
• [ php-wasm ] Add xdebug shared extension to @php-wasm/node JSPI #2248
• Playground CLI: Use Blueprints v2 #2281

In order to keep things moving, we merged #2231 while it still had some rough edges. This is an issue to follow up on those:

• Find way to stop skipping run-cli tests
• Find way to use normal test setup for test-built-npm-packages
• Fix Asyncify sqlite tests so we can stop skipping them
• Complete multi-worker support with Asyncify
• Consider using bootRequestHandler() instead of bootWordPress() after booting the first worker.
• Make sure error logs are collected from the correct worker and/or consolidated when needed.
• Consider alternate load balancing approach, perhaps "a moving average of total idle time per minute as a decent proxy metric"
• Look at unifying multi-worker management with PHP process management. We don't want to have to be aware of where or how PHP is running. Consider whether php-fpm or something like it is appropriate.
• Drop topOfStack testing from php-asyncify-sqlite3.spec.ts
• Use native fcntl for fcntl-style locks.
• Implement the two missing file-lock-manager test cases
• Support sharing flock-based locks with forked processes.
• Remove unnecessary nativeNodeModulesPlugin from php-wasm/node/build.js
• Reconcile js_getpid() with facilities we ship in the base emscripten js library. If the two don't have much overlap, let's at least acknowledge the other one in a comment. There may be some overlap, though, as proc_open can spawn a php process.
• Return an error from get_native_path_from_vfs_path() if no native path is found and make sure callers handle that.
• Make sure get_vfs_path_from_fd() doesn't abuse or mistakenly use a native /proc dir if exposed by useHostFilesystem().
• For node-es-module-loader ?url import handling, grep by /@fs and see if there's any impact on matching of vite paths with .href vs .pathname. If not, we're good.
• Look at the getpid() override and the proc_open() process ID management and make sure there are no conflicts between them.
• Confirm safety of node lookup in locking.is_path_to_shared_fs(). Do not assume successful lookup.
• Should --experimentalMultiWorker be a hidden CLI option?
• Remove duplicates from ASYNCIFY_IMPORTS lists
• Run performance test of multi-worker setup
• Make fcntl and flock overrides conditional on existence of PHPLoader.fileLockManager

Relevant but out of scope

• Consider how to stop GH actions killing tests without using so many runners to split unit tests

TODO: Add more items from #2231 review comments

[adamziel]

Complete multi-worker support with Asyncify

This is the most important item here

[brandonpayton]

This is the most important item here

@adamziel 👍 I've been working on this today.

There were some functions missing from the ASYNCIFY_IMPORTS list, but there is a "null or function signature mismatch" error related to fd_close.

This evening, I fought with the DWARF path mapping for the /emsdk/emscripten path to enable debugging the libc portions of the stack. With that work, other emscripten DWARF paths were mapped correctly, but the /emsdk path is not affected. It might be because the info is from a prebuilt library like libz. 🤔

Either way, this is not required for this troubleshooting. It just looked like a quick win, and it would be good to be able to step into all source code for a debug build.

In the morning, I'll post a draft PR with the work so far and focus on addressing the signature mismatch error.