eth: every gas estimate is sent to the alternative provider, burning its rate-limit quota

[cranycrane]

Problem

EthereumTypeEstimateGas (bchain/coins/eth/ethrpc.go:1961, provider block at ~1985) routes every eth_estimateGas request to the alternative send-tx provider's urls[0] whenever *_ALTERNATIVE_SENDTX_URLS is configured — the guard is a bare nil-check, with a fresh RPC client dialed per call and a silent fallback to the primary backend on error (the provider error is discarded, so the problem is invisible in logs).

This is the same pathology as #1627, one function further down the file: the routing is unconditional, but almost no request needs the provider. The path is hit from the websocket estimateFee handler (server/websocket.go:1124) — i.e. every time a user sits on a wallet's send screen, with estimates re-firing as the amount is edited.

Example

• An ETH instance has ETH_ALTERNATIVE_SENDTX_URLS pointed at a private relay whose API key allows ~10 req/s.
• Alice types an amount into the send form. The wallet requests a gas estimate; Blockbook sends eth_estimateGas to the relay, although Alice never submitted a private transaction — the primary backend could have answered, and the relay call spends quota for nothing.
• With ~50 users composing transactions concurrently, the relay quota is exhausted → 429s → every estimate pays a wasted round-trip (dial + call + rejection) before the primary answers, making the send form slower for everyone.
• Bob, who actually just broadcast a private transaction through the relay, is the one harmed: the quota his nonce lookup and next broadcast need was burned by everyone else's gas estimates.

Why not simply "never use the provider for estimates"

There is one legitimate case: a sender with a private transaction still pending. If Bob composes a second transaction that builds on the pending state of his first (private) one, only the relay sees that pending transaction and can estimate correctly — the primary backend cannot.

Proposed fix

Reuse the gating introduced by #1628 for nonces: route eth_estimateGas through the alternative provider only for recent private senders (useForNonces / the recentSenders map), directed at the URL that accepted their send (nonceURL); all other requests go straight to the primary backend. Also stop discarding the provider error silently — log it at warning level like the nonce fallback does.

Related: #1627, #1628 (deliberately scoped the estimateGas path out).

🤖 Generated with Claude Code

[Metavcl]

The proposed gating makes sense. eth_estimateGas is primarily read traffic and should not consume a private relay’s quota unless the sender actually depends on relay-only pending state.

Along with recentSenders and nonceURL, it would be useful to reuse RPC clients and expose route-level metrics for primary calls, alternative-provider calls, fallbacks, and 429s. That would make this regression immediately visible if it returns.

For production, the normal estimate/read path could use BlockPI’s Ethereum RPC or a Dedicated Node, while the private relay remains isolated for recent private senders. This keeps the send screen responsive without burning relay capacity needed for real private transactions.

You can try BlockPI at https://blockpi.io/ with promo code buildwithblockpi.