You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
bitreq: Support multiple connections per origin #658
Client currently caches one AsyncConnection per ConnectionKey, so capacity limits cached origins rather than connections to one origin.
Consequently, concurrent requests to the same HTTP/1.1 server share one socket. Pipelining does not provide server-side concurrency with common servers such as Go's net/http, resulting in head-of-line blocking. This is causing serialized remote-storage reads in: lightningdevkit/orange-sdk#90
We should add support for a configurable max_connections_per_origin, with connections selected by availability/load. The existing cache capacity could remain a separate limit on cached origins.
Pipelining does not provide server-side concurrency with common servers such as Go's net/http, resulting in head-of-line blocking. This is causing serialized remote-storage reads in: lightningdevkit/orange-sdk#90
That shouldn't matter here, though. For VSS, the server-side processing of a single request should be a few milliseconds, so the pipelining should still help a lot.
That shouldn't matter here, though. For VSS, the server-side processing of a single request should be a few milliseconds, so the pipelining should still help a lot.
But out-of-scope for this issue here, but in VSS we can pipeline reads can't pipeline writes/removes due to idempotency concerns. So IMO using multiple connections would help a lot here to avoid head of line blocking, especially in cases where writes and reads are interleaved, as any write would be a serialization point/bottleneck.
Activity
[-]Support multiple connections per origin[/-][+]`bitreq`: Support multiple connections per origin[/+]TheBlueMatt commentedon Jul 10, 2026
That shouldn't matter here, though. For VSS, the server-side processing of a single request should be a few milliseconds, so the pipelining should still help a lot.
tnull commentedon Jul 11, 2026
But out-of-scope for this issue here, but in VSS we can pipeline reads can't pipeline writes/removes due to idempotency concerns. So IMO using multiple connections would help a lot here to avoid head of line blocking, especially in cases where writes and reads are interleaved, as any write would be a serialization point/bottleneck.
TheBlueMatt commentedon Jul 11, 2026
Yea I feel like we should treat non-pipelined requests when there's no available no-requests-pending connections as needing a fresh connection always.
jamillambert commentedon Sep 1, 2026
Issue moved to Forgejo: https://gitea.bitcoin.ninja/rust-bitcoin/corepc/issues/658