Losing context across elasticsearch requests

[drewtunes]

I'm using the latest elasticsearch client (v11.0.1) and am running into issues with losing context. With a small number of concurrent requests, it works fine. However, as the load increases and I issue many elasticsearch requests, the context starts getting lost.

I can easily reproduce this in a unit test by dispatching about 25 requests simultaneously and comparing a value in context before and after. The first ten will work, the next 15 won't.

I've been digging into the elasticsearch client code, trying to find why it loses context, but haven't been very successful yet. I assume it has something to do with connection pooling. Is there some way I can wrap my code and basically re-attach the known context when the elasticsearch request returns?

I've tried using my patched Q as the defer handler but am now using the callback in my test to eliminate that as a possibility.

[drewtunes]

I've now confirmed that if I send two concurrent requests, the first will succeed and the second will lose context. The elasticsearch client code calls process.nextTick and after this point the context is lost.

[drewtunes]

Elasticsearch is using the http client and reusing sockets (as I believe it should for a database) so my issue may be due to this: #71

[Qard]

Internally, most async callbacks are wrapped with namespace.bind(callback) to ensure the context continues. You can try using that manually in your code to figure out at what point it works, then you'll know that somewhere between the call and callback the context is lost.

Generally, as you said, these sort of issues trace back to some sort of connection pooling or user-mode queuing of some variety.