dimas-b commented on code in PR #1517:
URL: https://github.com/apache/polaris/pull/1517#discussion_r2074339324
##########
extension/persistence/relational-jdbc/src/main/java/org/apache/polaris/extension/persistence/relational/jdbc/DatasourceOperations.java:
##########
@@ -173,23 +190,82 @@ public int executeUpdate(String query) throws
SQLException {
* @throws SQLException : Exception caught during transaction execution.
*/
public void runWithinTransaction(TransactionCallback callback) throws
SQLException {
- try (Connection connection = borrowConnection()) {
- boolean autoCommit = connection.getAutoCommit();
- connection.setAutoCommit(false);
- boolean success = false;
+ withRetries(
+ () -> {
+ try (Connection connection = borrowConnection()) {
+ boolean autoCommit = connection.getAutoCommit();
+ boolean success = false;
+ connection.setAutoCommit(false);
+ try {
+ try (Statement statement = connection.createStatement()) {
+ success = callback.execute(statement);
+ }
+ } finally {
+ if (success) {
+ connection.commit();
+ } else {
+ connection.rollback();
+ }
+ connection.setAutoCommit(autoCommit);
+ }
+ }
+ return null;
+ });
+ }
+
+ private boolean isRetryable(SQLException e) {
+ String sqlState = e.getSQLState();
+
+ if (sqlState != null) {
+ return sqlState.equals(DEADLOCK_SQL_CODE)
+ || // Deadlock detected
+ sqlState.equals(SERIALIZATION_FAILURE_SQL_CODE); // Serialization
failure
+ }
+
+ // Additionally, one might check for specific error messages or other
conditions
+ return e.getMessage().contains("connection refused")
+ || e.getMessage().contains("connection reset");
+ }
+
+ public <T> T withRetries(Operation<T> operation) throws SQLException {
+ int attempts = 0;
+ // maximum number of retries.
+ int maxAttempts = relationalJdbcConfiguration.maxRetries().orElse(1);
+ // How long we should try, since the first attempt.
+ long maxDuration =
relationalJdbcConfiguration.maxDurationInMs().orElse(100L);
+ // How long to wait before first failure.
+ long delay = relationalJdbcConfiguration.initialDelayInMs().orElse(100L);
+
+ // maximum time we will retry till.
+ long maxRetryTime = Instant.now().toEpochMilli() + maxDuration;
+
+ while (attempts < maxAttempts) {
try {
- try (Statement statement = connection.createStatement()) {
- success = callback.execute(statement);
+ return operation.execute();
+ } catch (SQLException e) {
+ attempts++;
+ long timeLeft = Math.max((maxRetryTime -
Instant.now().toEpochMilli()), 0L);
+ if (attempts >= maxAttempts || !isRetryable(e) || timeLeft == 0) {
+ throw e;
}
- } finally {
- if (success) {
- connection.commit();
- } else {
- connection.rollback();
+ // Add jitter
+ long timeToSleep = Math.min(timeLeft, delay + (long)
(random.nextDouble() * 0.2 * delay));
Review Comment:
So this formula gives jitter in the range `[0, 0.2 * delay]`. This is too
narrow IMHO. The purpose of jitter is to minimize the chance of secondary
collisions. I'd use the jitter range of `(0, delay]`. Please be careful with
`0` sleep time.
I personally do not see value in having a fixed component (`delay`) in the
sleep time. Merely delaying an operation is not helpful and we have not way of
predicting when remote transactions are going to finish.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
