doh: keep pooled conns shorter-lived than server idle-timeouts; h2 PING health-checks#242
Open
ShiroiKuma0 wants to merge 1 commit into
Open
doh: keep pooled conns shorter-lived than server idle-timeouts; h2 PING health-checks#242ShiroiKuma0 wants to merge 1 commit into
ShiroiKuma0 wants to merge 1 commit into
Conversation
…NG health-checks Some resolvers close idle DoH connections well under 30s (Quad9 measured at <=30s from the same LAN; Cloudflare survives >240s). With the 3m IdleConnTimeout, any DNS lull left only dead connections in the pool and subsequent queries were written into them, hanging ~10s and surfacing as 'Post ...: unexpected EOF' / http-status 502 bursts (maxEOFTries retries often just picked the next corpse). Downstream this expired alg mappings (realips([])), killed established flows of allowed apps, and failed the OS connectivity probes routed through the tunnel. - IdleConnTimeout 3m -> 10s: never reuse a connection older than the shortest observed server-side idle window. - http2.ConfigureTransports + ReadIdleTimeout/PingTimeout: actively evict half-dead h2 connections via PING instead of losing a query.
ignoramous
reviewed
Jul 18, 2026
| // subsequent queries were written into them — hanging ~10s and dying with | ||
| // "unexpected EOF" / http-status 502 bursts (maxEOFTries retries often just pick the next | ||
| // corpse). Keep pooled connections shorter-lived than any observed server-side idle window. | ||
| IdleConnTimeout: 10 * time.Second, |
Contributor
There was a problem hiding this comment.
10s seems too low. Perhaps this could be et to 30s instead?
ignoramous
reviewed
Jul 18, 2026
| // Health-check h2 connections with PING frames so half-dead ones are evicted instead of | ||
| // eating a query first (net/http does not do this by default). | ||
| if t2, err := http2.ConfigureTransports(tr); err == nil && t2 != nil { | ||
| t2.ReadIdleTimeout = 10 * time.Second |
Contributor
There was a problem hiding this comment.
Not sure what these are required for. Just the 30s idle timeout seems sufficient...?
ignoramous
requested changes
Jul 18, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #241.
h2()pools DoH connections for 3 minutes (IdleConnTimeout) with no HTTP/2 keep-alive health-check. Several resolvers close idle DoH connections far sooner — measured from a fixed network, same day, reusing a TLS connection after an idle period:With Quad9 (or any resolver whose idle timeout is under 3 minutes), every DNS lull longer than the server's idle window leaves the pool full of dead connections. The next queries are written into them, hang ~8–10 s, and surface as
Post "…": unexpected EOForhttp-status: 502;maxEOFTriesoften just picks the next dead pooled connection, so bursts of parallel queries fail together. Downstream on Android this expires alg mappings (realips([])), killing established flows of allowed apps, and fails the OS connectivity probes routed through the tunnel.Change
Two adjustments in
h2():IdleConnTimeout3m → 10s — never keep a pooled connection longer than the shortest observed server-side idle window.http2.ConfigureTransports+ReadIdleTimeout = 10s/PingTimeout = 5s— actively evict half-dead h2 connections via PING instead of losing a query to them.golang.org/x/netis already a module dependency, so the only new import isgolang.org/x/net/http2.Testing
Built from source and run on-device (Android, v055x/v055y engines). Same device, same resolver (Quad9), driven with worst-case traffic — bursts of 10 parallel uncached queries after guaranteed > 30 s idle gaps, ~40 cycles over 33 min:
retrier: read … err: EOFwith no query error following), roughly one per idle cycle.One caveat: with the PING enabled I saw a single rare
stream error: stream ID N; PROTOCOL_ERRORreset taking out one burst on one connection (once in ~40 cycles; recovered on redial). If that turns out to be an interaction with the custom dialer, theIdleConnTimeoutreduction alone already removes the corpse-pool state and may be the safer minimal change.