Feature Request: Support Worker Threads using Sharp

[metabench]

@lovell While I have read how Sharp itself would not benefit from worker threads in #1297, I am using Sharp alongside some CPU-intensive JS programming that certainly would benefit from being in a worker thread.

nodejs/node#21481 explains the issue, and at the bottom is a reference to how it was solved in node-lzo

[lovell]

Hello, sharp uses nan's NAN_MODULE_INIT and this feels like something that might be better added in that module so many native modules can benefit, especially as the NODE_MODULE_INIT macro is only available in Node 10.7.0+.

Having said that I'd be happy to accept a PR that does this as long as it remains fairly simple and has tests.

The lzo example does not appear to use libuv worker threads where sharp does so may not be a complete enough example.

The "context-aware addon carries with it the responsibility of carefully managing global static data" warning of https://nodejs.org/api/addons.html#addons_context_aware_addons is relevant here too.

[metabench]

Is it worth still using nan or shifting to N-API?

[lovell]

Please see #1282 for N-API.

[nstepien]

FYI for how iltorb does it: https://github.com/MayhemYDG/iltorb/blob/master/src/iltorb.cc#L7-L19

[lovell]

There's a discussion about adding this logic to nan at nodejs/nan#844

[vinerz]

+1. I am still suffering from some non-heap leakage at heroku for over a year now.
Would workers be a benefit for me as they eventually die?
I'm currently spawning a new child_process, which fixes the leak, but the overhead is unbelievable.
The request time went from 140ms to ~1500ms after this workaround.

EDIT: jemalloc in fact did the job and fixed the memory issues. Hooray!
For anyone wondering how to do that in Heroku, use the jemalloc buildpack.

[reutopiaer]

I can user sharp in Worker Threads （sharp:v0.22.1,node:v10.15.3).
But when I require sharp in mainThread and workerThread both, I still get error: "Module did not self-register."

[lovell]

Commit 946d3c8 makes the switch to use the new NAN_MODULE_WORKER_ENABLED macro. This will be in v0.23.0.

I've flagged this as experimental in the changelog as although both sharp and libvips are very much thread-aware, this probably requires testing at scale before use in production.

[lovell]

sharp v0.23.0 is now available.

[MarcusCemes]

Hi,
Would anybody know by any chance if worker threads have their own libuv threadpool? In my tests, worker threads share the exact same sharp job queue, which is limited to 4 concurrent images (without setting UV_THREADPOOL_SIZE) and can even be queried from the main thread (sharp.counters()). Is there any way to give each thread its own set of processing threads? I have a code sample if someone is interested.

My current solution is to spawn more Node.js process, but threads would be lighter. I don't want to bug my user to set UV_THREADPOOL_SIZE either for a CLI tool.

[dnlup]

Hi,
Would anybody know by any chance if worker threads have their own libuv threadpool? In my tests, worker threads share the exact same sharp job queue, which is limited to 4 concurrent images (without setting UV_THREADPOOL_SIZE) and can even be queried from the main thread (sharp.counters()). Is there any way to give each thread its own set of processing threads? I have a code sample if someone is interested.

My current solution is to spawn more Node.js process, but threads would be lighter. I don't want to bug my user to set UV_THREADPOOL_SIZE either for a CLI tool.

@MarcusCemes worker threads all share the same thread pool, you can't much about it. You could try to use N workers, and in each worker set sharp concurrency to 1, but I don't think the performance would be better.