[
https://issues.apache.org/jira/browse/ARTEMIS-5001?focusedWorklogId=932295&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-932295
]
ASF GitHub Bot logged work on ARTEMIS-5001:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 28/Aug/24 23:23
Start Date: 28/Aug/24 23:23
Worklog Time Spent: 10m
Work Description: clebertsuconic commented on code in PR #5172:
URL: https://github.com/apache/activemq-artemis/pull/5172#discussion_r1735386767
##########
artemis-server/src/main/java/org/apache/activemq/artemis/core/persistence/impl/journal/OperationContextImpl.java:
##########
@@ -165,56 +185,74 @@ public synchronized void replicationDone() {
@Override
public void executeOnCompletion(IOCallback runnable) {
- executeOnCompletion(runnable, false);
+ executeOnCompletion(runnable, OperationConsistencyLevel.FULL);
}
@Override
- public void executeOnCompletion(final IOCallback completion, final boolean
storeOnly) {
+ public void executeOnCompletion(final IOCallback completion, final
OperationConsistencyLevel consistencyLevel) {
boolean executeNow = false;
synchronized (this) {
if (errorCode == -1) {
final long storeLined = STORE_LINEUP_UPDATER.get(this);
final long pageLined = PAGE_LINEUP_UPDATER.get(this);
final long replicationLined = REPLICATION_LINEUP_UPDATER.get(this);
- if (storeOnly) {
- if (storeOnlyTasks == null) {
- storeOnlyTasks = new LinkedList<>();
- }
- } else {
- if (tasks == null) {
- tasks = new LinkedList<>();
- minimalReplicated = replicationLined;
- minimalStore = storeLined;
- minimalPage = pageLined;
- }
- }
- // On this case, we can just execute the context directly
-
- if (replicationLined == replicated && storeLined == stored &&
pageLined == paged) {
- // We want to avoid the executor if everything is complete...
- // However, we can't execute the context if there are
executions pending
- // We need to use the executor on this case
- if (EXECUTORS_PENDING_UPDATER.get(this) == 0) {
- // No need to use an executor here or a context switch
- // there are no actions pending.. hence we can just execute
the task directly on the same thread
- executeNow = true;
- } else {
- execute(completion);
- }
- } else {
- if (storeOnly) {
- if (storeLined == stored &&
EXECUTORS_PENDING_UPDATER.get(this) == 0) {
- executeNow = true;
+ switch (consistencyLevel) {
+ case STORAGE:
+ if (storeOnlyTasks == null) {
+ storeOnlyTasks = new LinkedList<>();
+ }
+ if (storeLined == stored) {
+ if (hasNoPendingExecution()) {
+ executeNow = true;
+ } else {
+ execute(completion);
+ }
} else {
- assert !storeOnlyTasks.isEmpty() ?
storeOnlyTasks.peekLast().storeLined <= storeLined : true;
storeOnlyTasks.add(new StoreOnlyTaskHolder(completion,
storeLined));
}
- } else {
- // ensure total ordering
- assert validateTasksAdd(storeLined, replicationLined,
pageLined);
- tasks.add(new TaskHolder(completion, storeLined,
replicationLined, pageLined));
- }
+ break;
+
+ case IGNORE_REPLICATION:
+ if (ignoreReplicationTasks == null) {
+ ignoreReplicationTasks = new LinkedList<>();
+ }
+
+ if (storeLined == stored && pageLined == paged) {
+ if (hasNoPendingExecution()) {
+ // No need to use an executor here or a context switch
+ // there are no actions pending.. hence we can just
execute the task directly on the same thread
+ executeNow = true;
+ } else {
+ execute(completion);
+ }
+ } else {
+ ignoreReplicationTasks.add(new TaskHolder(completion,
storeLined, replicationLined, pageLined));
Review Comment:
I thought it was not needed... ignore replication is only used on Mirroring
(although you never know if we may expand usage). Storage is more used..
but I'm adding it just in case.
Issue Time Tracking
-------------------
Worklog Id: (was: 932295)
Time Spent: 5h (was: 4h 50m)
> Add configuration option to relax syncs journal replication for Mirror Target
> -----------------------------------------------------------------------------
>
> Key: ARTEMIS-5001
> URL: https://issues.apache.org/jira/browse/ARTEMIS-5001
> Project: ActiveMQ Artemis
> Issue Type: Improvement
> Affects Versions: 2.37.0
> Reporter: Clebert Suconic
> Assignee: Clebert Suconic
> Priority: Major
> Fix For: 2.38.0
>
> Time Spent: 5h
> Remaining Estimate: 0h
>
> When I worked on AMQP Mirror I did not actually envision being used with
> journal replication. I actually thought more about adding multiple mirrored
> options instead.
> However an user reported me that when using mirror and journal replication
> combined, the sends could take a lot longer to happen (some normal latency)
> and the acks would eventually be missed.
> I should add an option to ignore the replication for the Mirror Target.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
For further information, visit: https://activemq.apache.org/contact
