[
https://issues.apache.org/jira/browse/GEODE-8240?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17145952#comment-17145952
]
ASF GitHub Bot commented on GEODE-8240:
---------------------------------------
Bill commented on a change in pull request #5269:
URL: https://github.com/apache/geode/pull/5269#discussion_r442241602
##########
File path:
geode-membership/src/main/java/org/apache/geode/distributed/internal/membership/gms/GMSMemberData.java
##########
@@ -220,12 +220,16 @@ public InetAddress getInetAddress() {
@Override
public short getVersionOrdinal() {
- return this.versionObj.ordinal();
+ return versionOrdinal;
}
@Override
public Version getVersion() {
- return versionObj;
+ try {
+ return Version.fromOrdinal(versionOrdinal);
+ } catch (final UnsupportedSerializationVersionException e) {
Review comment:
**This right here** is a big fat problem. We are silently throwing away
information when the version ordinal represents an unknown version. One
approach is to let the exception escape (as a checked exception.) But that
entails fixing the 53 callers of this method. An alternative would be to
"tunnel" the checked exception through an unchecked one. Lastly we could leave
it as-is and try to go around and analyze whether or not the places calling
this can survive this loss of information. That would be bad enough but what's
worse is that we'd be introducing a maintenance trap.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> View has old locator version number after rolling upgrade
> ---------------------------------------------------------
>
> Key: GEODE-8240
> URL: https://issues.apache.org/jira/browse/GEODE-8240
> Project: Geode
> Issue Type: Bug
> Components: client/server, membership
> Reporter: Ernest Burghardt
> Assignee: Bill Burcham
> Priority: Major
>
> as shown in [https://github.com/apache/geode/pull/5224]
> locator upgrade from version 1.12.0 doesn't seem to occur
> {{testRollServersOnPartitionedRegion_dataserializable}} failure results:
> Expecting:
> <"Member Count : 3
> Name | Id
> ---- |
> -------------------------------------------------------------------------------
> vm2 | 127.0.0.1(vm2:35019:locator)<ec><v17>:41000(version:GEODE 1.12.0)
> [Coordinator]
> vm0 | 10.0.0.111(vm0:35025)<v27>:41001
> vm1 | 10.0.0.111(vm1:35030)<v29>:41002
> ">
> not to contain:
> <"1.12.0">
> This problem was introduced in 1.12.0 and is present in all lines derived
> from that one, including 9.10, 1.13, and current develop/1.14
> What's actually happening is that the locator _is_ upgraded to a newer
> version. It joins with an older coordinator (that's running e.g. 1.12.0) and
> that coordinator produces a view showing the new locator/member as running
> the same version, in this case 1.12.0, as the coordinator.
> Eventually, all locators will be upgraded. But the view carries the incorrect
> version indication.
> The root cause seems to be that when {{GMSMemberData.setVersionObject(short
> versionOrdinal)}} sees a version ordinal that is unknown, i.e. a version
> ordinal corresponding to a new line of development: 1.13, 1.14, … that method
> throws away that version ordinal and replaces it with the one for the 1.12
> line.
> Since the current {{support/1.13}} and {{develop}} branches have the bug
> upgrading a current 1.13 to 1.14 or a current development/1.14 to 1.15 would
> exhibit the same behavior (locator apparently stuck at the older version in
> the view.)
> Ramifications of this incorrect version indication in the view are TBD.
> Whether or not this situation resolves itself after _another_ round of
> restarts is TBD.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
