[
https://issues.apache.org/jira/browse/ARTEMIS-1474?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16212686#comment-16212686
]
ASF GitHub Bot commented on ARTEMIS-1474:
-----------------------------------------
Github user franz1981 commented on a diff in the pull request:
https://github.com/apache/activemq-artemis/pull/1602#discussion_r145973529
--- Diff:
artemis-journal/src/main/java/org/apache/activemq/artemis/core/io/buffer/TimedBuffer.java
---
@@ -34,6 +34,25 @@
import org.apache.activemq.artemis.journal.ActiveMQJournalLogger;
public final class TimedBuffer {
+
+ /**
+ * Property name to set the percentage of error allowed while expiring
the flush {@code timeout} to happen:
+ * it can assume any positive value from {@code 0} to {@link
Integer#MAX_VALUE}.
+ * <p>
+ * By default it is {@link #DEFAULT_TIMEOUT_ERROR_PERCENTAGE} more than
the configured {@code timeout}.
+ */
+ public static final String JOURNAL_TIMEOUT_ERROR_PROPERTY_NAME =
"journal.timeout.error";
+ public static final int DEFAULT_TIMEOUT_ERROR_PERCENTAGE = 50;
+ private static final double MAX_TIMEOUT_ERROR;
+
+ static {
+ final int errorPercentage =
Integer.getInteger(JOURNAL_TIMEOUT_ERROR_PROPERTY_NAME,
DEFAULT_TIMEOUT_ERROR_PERCENTAGE);
+ if (errorPercentage < 0) {
+ throw new RuntimeException("The sleep error percentage must be >=
0");
+ }
+ MAX_TIMEOUT_ERROR = 1 + (errorPercentage / 100);
--- End diff --
In the old code was indeed:
```
// I'm letting the real time to be up to 50% than the
requested sleep.
if (realTimeSleep > timeout * 1.5) {
failedChecks++;
}
```
> TimedBuffer need sleep error detection measured against the expected timeout
> ----------------------------------------------------------------------------
>
> Key: ARTEMIS-1474
> URL: https://issues.apache.org/jira/browse/ARTEMIS-1474
> Project: ActiveMQ Artemis
> Issue Type: Improvement
> Reporter: Francesco Nigro
> Assignee: Francesco Nigro
> Priority: Minor
>
> The current TimedBuffer implementation evaluate the sleep accuracy measuring
> the error against the elapsed sleep time while the original version was using
> the error related to the overall expected timeout: It would be better due to
> the sleep/parkNanos behaviour to maintain the original evaluation method, but
> using the improved discounted (with the last flush time) sleep time in order
> to not exeeding to delay flushes.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
