rpcclient: add context.Context support to RPC methods

[seeforschauer]

Problem

rpcclient methods (GetBlockCount, GetBlockHash, SendRawTransaction, etc.) don't accept context.Context. Callers have no way to cancel or timeout individual RPC calls. When a node is slow or unresponsive, GetBlockCount() blocks for up to 600s (10 retries x 60s HTTP timeout) with no way to interrupt it.

The root cause is in handleSendPostMessage — HTTP requests are created via http.NewRequest (without context), and the retry loop has no ctx.Done() check:

// infrastructure.go — retry loop, no cancellation
for i := 0; tries == 0 || i < tries; i++ {
    resp, err = httpClient.Do(req)  // 60s timeout each
    <-time.After(backoff)           // no select on ctx.Done()
}

I searched existing issues and PRs — #2450 and #2451 improve shutdown and dial timeout handling but don't address per-request context cancellation. No other open issues cover this.

Use case

We run a HAProxy health checker that polls BTC nodes via GetBlockCount(). HAProxy sends a TCP probe every N seconds; each probe spawns a goroutine calling GetBlockCount(). When a node hangs:

• Each probe creates a new goroutine blocked in GetBlockCount for ~600s
• New probes arrive faster than old ones complete
• Goroutines accumulate — we observed 117 leaked goroutines on staging

We worked around this with a singleflight pattern + context timeout at our layer, but the proper fix belongs in the library.

Proposed approach

Non-breaking: add *WithContext variants alongside existing methods (same pattern as database/sql).

ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
count, err := client.GetBlockCountWithContext(ctx)

Minimal changes to infrastructure.go:

1. jsonRequest — add ctx context.Context field
2. handleSendPostMessage — http.NewRequestWithContext(jReq.ctx, ...) + ctx.Done() in retry loop
3. Client.SendCmdCtx(ctx, cmd) — new method, passes context through
4. GetBlockCountWithContext(ctx) — first method using the plumbing

Builds on #2450 and #2451. Remaining methods can follow in incremental PRs. ~80-100 lines total.

Happy to submit a PR if maintainers agree with the approach.

Related

• rpcclient: make shutdown interrupt in-flight POSTs #2451 — shutdown interrupts in-flight POSTs (same area)
• rpcclient: ensure http dial respects timeout #2450 — dial respects timeout (merged)
• rpcclient.Client doesn't reuse network connections #1323 — connection reuse (related transport issue)

[seeforschauer]

I'd like to take this on — planning to submit a PR in the coming days with context-aware variants of the RPC methods (starting with the most commonly used ones like GetBlockCount, GetBlock, GetBlockHash).

The approach: add a ctx field to jsonRequest, wire it through handleSendPostMessage via http.NewRequestWithContext, and expose *WithContext method variants. Non-breaking — existing methods keep their signatures.