[
https://issues.apache.org/jira/browse/GEODE-2694?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15965147#comment-15965147
]
Eric Shu commented on GEODE-2694:
---------------------------------
There are two places that entry will be destroyed when recovered_from_disk bit
is set to true.
After processing a full gii, all entries with the recovered_from_disk bit set
to true will be destroyed, possibly try to remove the unfinished operations.
final void finishInitializeOwner(LocalRegion drs, GIIStatus giiStatus) {
if (isReadyForRecovery()) {
// this.scheduleCompaction();
if (GIIStatus.didFullGII(giiStatus)) {
--> destroyRemainingRecoveredEntries(drs);
} else if (GIIStatus.didDeltaGII(giiStatus)) {
// TODO: not sure if we should destroy old tombstones for deltaGII
} else if (getRegionVersionVector() != null) {
destroyOldTomstones(drs);
}
releaseRecoveryData();
}
and here in InitialImageOperation.getFromOne
// review unfinished keys and remove untouched entries
if (this.region.getDataPolicy().withPersistence() &&
keysOfUnfinishedOps != null
&& !keysOfUnfinishedOps.isEmpty()) {
final DiskRegion dr = this.region.getDiskRegion();
assert dr != null;
for (Object key : keysOfUnfinishedOps) {
RegionEntry re = this.entries.getEntry(key);
if (re == null) {
continue;
}
if (logger.isTraceEnabled(LogMarker.GII)) {
logger.trace(LogMarker.GII, "Processing unfinished
operation:entry={}", re);
}
DiskEntry de = (DiskEntry) re;
synchronized (de) {
DiskId id = de.getDiskId();
if (id != null &&
EntryBits.isRecoveredFromDisk(id.getUserBits())) {
this.region.destroyRecoveredEntry(key);
if (isDebugEnabled) {
logger.debug("Deleted unfinished keys:key={}", key);
}
}
}
The bit set in dr.testIsRecoveredAndClear is clearly needed during finishing
gii.
During full gii, we could rewrite the entry to disk -- the optimization in the
following could be thought as the entry was rewritten to disk as they have the
same version tag.
// If the received entry and what we have in the cache
// actually are equal, keep don't put the received
// entry into the cache (this avoids writing a record to disk)
if (entriesEqual) {
continue;
}
Will add unit tests to verify the RECOVERED_FROM_DISK bit after GII.
> RECOVERED_FROM_DISK bit is cleared during gii, but should be restored if the
> recovered entry and gii entry has the same version tag
> -----------------------------------------------------------------------------------------------------------------------------------
>
> Key: GEODE-2694
> URL: https://issues.apache.org/jira/browse/GEODE-2694
> Project: Geode
> Issue Type: Bug
> Components: regions
> Reporter: Eric Shu
>
> Currently for all gii entries, product clears the RECOVERED_FROM_DISK bit for
> DiskEntry. However, if entry comes from gii has the same version as recovered
> entry, the RECOVERED_FROM_DISK bit should be restored but does not.
> {noformat}
> synchronized (re) { // fixes bug 41409
> if (dr.testIsRecoveredAndClear(re)) {
> wasRecovered = true;
> if (tmpValue == null) {
> tmpValue = entry.isLocalInvalid() ? Token.LOCAL_INVALID :
> Token.INVALID;
> }
> // Compare the version stamps, and if they are equal
> // we can skip adding the entry we receive as part of GII.
> VersionStamp stamp = re.getVersionStamp();
> boolean entriesEqual = stamp != null &&
> stamp.asVersionTag().equals(tag);
> // If the received entry and what we have in the cache
> // actually are equal, keep don't put the received
> // entry into the cache (this avoids writing a record to disk)
> if (entriesEqual) {
> continue;
> }
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
