[
https://issues.apache.org/jira/browse/TEPHRA-223?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15877027#comment-15877027
]
ASF GitHub Bot commented on TEPHRA-223:
---------------------------------------
Github user chtyim commented on a diff in the pull request:
https://github.com/apache/incubator-tephra/pull/37#discussion_r102348469
--- Diff:
tephra-core/src/main/java/org/apache/tephra/TransactionManager.java ---
@@ -1123,15 +1112,14 @@ private boolean doTruncateInvalidTxBefore(long
time) throws InvalidTruncateTimeE
}
// Find all invalid transactions earlier than truncateWp
- Set<Long> toTruncate = Sets.newHashSet();
- for (long wp : invalid) {
- // invalid list is sorted, hence can stop as soon as we reach a wp
>= truncateWp
- if (wp >= truncateWp) {
- break;
+ LongSet toTruncate = new LongArraySet();
+ for (long wp : invalidTxList.toRawList()) {
--- End diff --
Unfortunately using the normal for-each loop means there would be boxing.
To avoid boxing, you have to use the iterator directly
```java
LongIterator it = invalidTxList.toRawList().iterator();
while (it.hasNext()) {
long wp = it.nextLong();
...
}
```
> Transactions started after a snapshot restore can have incorrect invalid
> transaction list
> -----------------------------------------------------------------------------------------
>
> Key: TEPHRA-223
> URL: https://issues.apache.org/jira/browse/TEPHRA-223
> Project: Tephra
> Issue Type: Bug
> Components: core
> Affects Versions: 0.5.0, 0.6.5, 0.7.0, 0.8.0-incubating, 0.9.0-incubating,
> 0.10.0-incubating
> Reporter: Poorna Chandra
> Assignee: Poorna Chandra
> Fix For: 0.12.0-incubating
>
>
> Transaction Manager uses two datastructures to manage the invalid list - a
> {{List}} and an {{array}}. All updates to invalid list happen to the
> {{List}}, and the list is then sorted and copied over to the {{array}}. This
> is an optimization done so as to not sort the invalid list every time a new
> transaction is created.
> However during a snapshot restore after the Transaction Manager starts up,
> the invalid list gets correctly restored to the {{List}} but does not get
> copied over to the {{array}}. This will lead to transactions started right
> after a snapshot restore to have empty invalid list. This will make invalid
> data to become visible to those transactions.
> Since the {{List}} still contains the right invalid list, any update to the
> invalid list - like invalidating a transaction, will restore the {{array}}
> back to a good state.
> Also compactions will still remove invalid data as expected since new
> snapshots are searialized using the {{List}}
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
