Classify Arbitrum rpc server errors (#15488)

* Add service errors * Add default service errors * Fix tests * Update giant-eels-jump.md * Update errors.go
smartcontractkit · Dec 12, 2024 · 52f364a · 52f364a
1 parent acc3846
commit 52f364a
Show file tree

Hide file tree

Showing 3 changed files with 33 additions and 2 deletions.
diff --git a/.changeset/giant-eels-jump.md b/.changeset/giant-eels-jump.md
@@ -0,0 +1,5 @@
+---
+"chainlink": patch
+---
+
+Add error handling for Arbitrum RPC server timeouts. #added
diff --git a/core/chains/evm/client/errors.go b/core/chains/evm/client/errors.go
@@ -64,6 +64,7 @@ const (
 	ServiceUnavailable
 	TerminallyStuck
 	TooManyResults
+	ServiceTimeout
 )
 
 type ClientErrors map[int]*regexp.Regexp
@@ -160,7 +161,8 @@ var arbitrum = ClientErrors{
 	Fatal:                 arbitrumFatal,
 	L2FeeTooLow:           regexp.MustCompile(`(: |^)max fee per gas less than block base fee(:|$)`),
 	L2Full:                regexp.MustCompile(`(: |^)(queue full|sequencer pending tx pool full, please try again)(:|$)`),
-	ServiceUnavailable:    regexp.MustCompile(`(: |^)502 Bad Gateway: [\s\S]*$|network is unreachable|i/o timeout`),
+	ServiceUnavailable:    regexp.MustCompile(`(: |^)502 Bad Gateway: [\s\S]*$|network is unreachable|i/o timeout|(: |^)503 Service Temporarily Unavailable(:|$)`),
+	ServiceTimeout:        regexp.MustCompile(`(: |^)408 Request Timeout(:|$)`),
 }
 
 // Treasure
@@ -398,6 +400,11 @@ func (s *SendError) IsServiceUnavailable(configErrors *ClientErrors) bool {
 	return s.is(ServiceUnavailable, configErrors) || pkgerrors.Is(s.err, commonclient.ErroringNodeError)
 }
 
+// IsServiceTimeout indicates if the error was caused by a service timeout
+func (s *SendError) IsServiceTimeout(configErrors *ClientErrors) bool {
+	return s.is(ServiceTimeout, configErrors)
+}
+
 // IsTerminallyStuck indicates if a transaction was stuck without any chance of inclusion
 func (s *SendError) IsTerminallyStuckConfigError(configErrors *ClientErrors) bool {
 	return s.is(TerminallyStuck, configErrors)
@@ -619,6 +626,10 @@ func ClassifySendError(err error, clientErrors config.ClientErrors, lggr logger.
 		lggr.Errorw(fmt.Sprintf("service unavailable while sending transaction %x", tx.Hash()), "err", sendError, "etx", tx)
 		return commonclient.Retryable
 	}
+	if sendError.IsServiceTimeout(configErrors) {
+		lggr.Errorw(fmt.Sprintf("service timed out while sending transaction %x", tx.Hash()), "err", sendError, "etx", tx)
+		return commonclient.Retryable
+	}
 	if sendError.IsTimeout() {
 		lggr.Errorw(fmt.Sprintf("timeout while sending transaction %x", tx.Hash()), "err", sendError, "etx", tx)
 		return commonclient.Retryable
@@ -666,7 +677,7 @@ var drpc = ClientErrors{
 
 // Linkpool, Blockdaemon, and Chainstack all return "request timed out" if the log results are too large for them to process
 var defaultClient = ClientErrors{
-	TooManyResults: regexp.MustCompile(`request timed out`),
+	TooManyResults: regexp.MustCompile(`request timed out|408 Request Timed Out`),
 }
 
 // JSON-RPC error codes which can indicate a refusal of the server to process an eth_getLogs request because the result set is too large

diff --git a/core/chains/evm/client/errors_test.go b/core/chains/evm/client/errors_test.go
@@ -245,6 +245,7 @@ func Test_Eth_Errors(t *testing.T) {
 			{"network is unreachable", true, "Arbitrum"},
 			{"client error service unavailable", true, "tomlConfig"},
 			{"[Request ID: 825608a8-fd8a-4b5b-aea7-92999509306d] Error invoking RPC: [Request ID: 825608a8-fd8a-4b5b-aea7-92999509306d] Transaction execution returns a null value for transaction", true, "hedera"},
+			{"call failed: 503 Service Temporarily Unavailable: <html>\r\n<head><title>503 Service Temporarily Unavailable</title></head>\r\n<body>\r\n<center><h1>503 Service Temporarily Unavailable</h1></center>\r\n</body>\r\n</html>\r\n", true, "Arbitrum"},
 		}
 		for _, test := range tests {
 			err = evmclient.NewSendErrorS(test.message)
@@ -260,6 +261,20 @@ func Test_Eth_Errors(t *testing.T) {
 		}
 	})
 
+	t.Run("IsServiceTimeout", func(t *testing.T) {
+		tests := []errorCase{
+			{"call failed: 408 Request Timeout: {", true, "Arbitrum"},
+			{"408 Request Timeout: {\"id\":303,\"jsonrpc\":\"2.0\",\"error\":{\"code\\\":-32009,\\\"message\\\":\\\"request timeout\\\"}}\",\"errVerbose\":\"408 Request Timeout:\n", true, "Arbitrum"},
+			{"request timeout", false, "tomlConfig"},
+		}
+		for _, test := range tests {
+			err = evmclient.NewSendErrorS(test.message)
+			assert.Equal(t, err.IsServiceTimeout(clientErrors), test.expect)
+			err = newSendErrorWrapped(test.message)
+			assert.Equal(t, err.IsServiceTimeout(clientErrors), test.expect)
+		}
+	})
+
 	t.Run("IsTxFeeExceedsCap", func(t *testing.T) {
 		tests := []errorCase{
 			{"tx fee (1.10 ether) exceeds the configured cap (1.00 ether)", true, "geth"},