Merge pull request #1983 from LiskHQ/1979-missing-error-handling-in-c…

…oordinator

Missing error handling in coordinator

docker/example.env

@@ -32,7 +32,7 @@
 # CLIENT_INSTANTIATION_RETRY_INTERVAL=1
 # CLIENT_ALIVE_ASSUMPTION_TIME=5000
 # HEARTBEAT_ACK_MAX_WAIT_TIME=1000
-# ENDPOINT_INVOKE_MAX_RETRIES=5
+# ENDPOINT_INVOKE_MAX_RETRIES=3
 # ENDPOINT_INVOKE_RETRY_DELAY=10
 # CONNECTOR_EXIT_DELAY_IN_HOURS=0
 

docs/antora/modules/ROOT/pages/configuration/index.adoc

@@ -428,7 +428,7 @@ By default, it is set to `1000`.
 | `ENDPOINT_INVOKE_MAX_RETRIES`
 | number
 | Maximum number of endpoint invocation request retries to the node.
-By default, it is set to `5`.
+By default, it is set to `3`.
 | 5
 
 | `ENDPOINT_INVOKE_RETRY_DELAY`

ecosystem.config.js

@@ -131,7 +131,7 @@ module.exports = {
 				// CLIENT_INSTANTIATION_RETRY_INTERVAL: 1,
 				// CLIENT_ALIVE_ASSUMPTION_TIME: 5 * 1000,
 				// HEARTBEAT_ACK_MAX_WAIT_TIME: 1000,
-				// ENDPOINT_INVOKE_MAX_RETRIES: 5,
+				// ENDPOINT_INVOKE_MAX_RETRIES: 3,
 				// ENDPOINT_INVOKE_RETRY_DELAY: 10,
 				// CONNECTOR_EXIT_DELAY_IN_HOURS: 0,
 				// JOB_INTERVAL_CACHE_CLEANUP: 0,

services/blockchain-connector/README.md

@@ -38,7 +38,7 @@ A list of the most commonly used environment variables is presented below:
 - `CLIENT_INSTANTIATION_RETRY_INTERVAL`: Retry interval (in milliseconds) to invoke instantiate API client when getApiClient is invoked. By default, it is set to 5.
 - `CLIENT_ALIVE_ASSUMPTION_TIME`: Interval (in milliseconds) for which the WS API Client is assumed to be alive since the last ping/pong success check. By default, it is set to 5000.
 - `HEARTBEAT_ACK_MAX_WAIT_TIME`: Maximum time (in milliseconds) within which the checkIsClientAlive's algorithm expects a corresponding `pong` for the `ping` sent to the WS server. By default, it is set to 1000.
-- `ENDPOINT_INVOKE_MAX_RETRIES`: Maximum number of endpoint invocation request retries to the node. By default, it is set to 1.
+- `ENDPOINT_INVOKE_MAX_RETRIES`: Maximum number of endpoint invocation request retries to the node. By default, it is set to 3.
 - `ENDPOINT_INVOKE_RETRY_DELAY`: Delay (in milliseconds) between each endpoint invocation request retry. By default, it is set to 10.
 - `CONNECTOR_EXIT_DELAY_IN_HOURS`: Delay (in hours) after which the blockchain-connector microservice exits. The service should restart automatically if deployed using Docker or PM2. To be removed eventually. To enable it, set it higher than `0`. By default, it is set to `0`.
 - `JOB_INTERVAL_CACHE_CLEANUP`: Job run interval to cleanup block cache. By default, it is set to 0.

services/blockchain-connector/config.js

@@ -120,7 +120,7 @@ config.apiClient = {
 		retryInterval: Number(process.env.CLIENT_INSTANTIATION_RETRY_INTERVAL) || 1, // in millisecs
 	},
 	request: {
-		maxRetries: Number(process.env.ENDPOINT_INVOKE_MAX_RETRIES) || 5,
+		maxRetries: Number(process.env.ENDPOINT_INVOKE_MAX_RETRIES) || 3,
 		retryDelay: Number(process.env.ENDPOINT_INVOKE_RETRY_DELAY) || 10, // in millisecs
 	},
 };

services/blockchain-connector/events/controller/blockchain.js

@@ -14,6 +14,7 @@
  *
  */
 const { Logger, Signals } = require('lisk-service-framework');
+const { MODULE_NAME_POS } = require('../../shared/sdk/constants/names');
 
 const { getBlockByID } = require('../../shared/sdk/endpoints');
 const { formatBlock: formatBlockFromFormatter } = require('../../shared/sdk/formatter');
@@ -92,12 +93,20 @@ const chainNewBlockController = async cb => {
 				transactions,
 			}),
 		);
+
+		// Reload validators cache on pos module transactions
+		if (transactions.some(t => t.module === MODULE_NAME_POS)) {
+			Signals.get('reloadAllPosValidators').dispatch();
+		}
 	};
 	Signals.get('chainNewBlock').add(chainNewBlockListener);
 };
 
 const chainDeleteBlockController = async cb => {
-	const chainDeleteBlockListener = async payload => cb(formatBlock(payload));
+	const chainDeleteBlockListener = async payload => {
+		cb(formatBlock(payload));
+		Signals.get('reloadAllPosValidators').dispatch();
+	};
 	Signals.get('chainDeleteBlock').add(chainDeleteBlockListener);
 };
 

services/blockchain-connector/shared/sdk/index.js

@@ -128,6 +128,7 @@ const init = async () => {
 	await getTokenInitializationFees();
 	await getRewardTokenID();
 	await getPosConstants();
+	await getAllPosValidators();
 
 	// Download the genesis block, if applicable
 	await getGenesisBlock().then(() => {

services/blockchain-connector/shared/sdk/pos.js

@@ -15,6 +15,7 @@
  */
 const {
 	Logger,
+	Signals,
 	Exceptions: { TimeoutException },
 } = require('lisk-service-framework');
 
@@ -27,6 +28,7 @@ const { getGenesisHeight } = require('./genesisBlock');
 const logger = Logger();
 
 let posModuleConstants;
+let allPosValidators;
 
 const getPosValidator = async address => {
 	try {
@@ -43,10 +45,16 @@ const getPosValidator = async address => {
 	}
 };
 
-const getAllPosValidators = async () => {
+const getAllPosValidators = async isForceReload => {
 	try {
-		const validators = await invokeEndpoint('pos_getAllValidators');
-		return validators;
+		if (!allPosValidators || isForceReload) {
+			const response = await invokeEndpoint('pos_getAllValidators');
+			if (Array.isArray(response)) {
+				allPosValidators = response;
+				logger.info(`Reloaded pos validators list. Validators count: ${allPosValidators.length}.`);
+			}
+		}
+		return allPosValidators;
 	} catch (err) {
 		if (err.message.includes(timeoutMessage)) {
 			throw new TimeoutException("Request timed out when calling 'getAllPosValidators'.");
@@ -56,6 +64,8 @@ const getAllPosValidators = async () => {
 	}
 };
 
+Signals.get('reloadAllPosValidators').add(() => getAllPosValidators(true));
+
 const getPosValidatorsByStake = async limit => {
 	try {
 		const validators = await invokeEndpoint('pos_getValidatorsByStake', { limit });