[GEODE-6614] WAN concurrency conflict checks are messed up

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Won't Fix
• Component/s: regions, wan
• Assignee: Unassigned
• Reporter: Bruce J Schuchardt
• Created: 08/Apr/19 21:25
• Updated: 11/Apr/19 22:40
• Resolved: 11/Apr/19 22:40

Description

Someone inserted code into concurrency conflict checks that treats all VersionTags as gateway tags if their distributed system ID is not the same as the RegionEntry being modified.  When that's the case the code checks timestamps and distributed system identifiers to resolve conflicts.  This is counter-intuitive for folks trying to debug WAN consistency issues and can cause odd decisions to be made, especially in client caches.

The revision that did this is so old it's not in the Apache github repo, but the code in question is

// bug #46223, an event received from a peer or a server may be from a different
// distributed system than the last modification made to this entry so we must
// perform a gateway conflict check
if (stamp != null && !tag.isAllowedByResolver()) {
  int stampDsId = stamp.getDistributedSystemId();
  int tagDsId = tag.getDistributedSystemId();

  if (stampDsId != 0 && stampDsId != tagDsId && stampDsId != -1) {
    StringBuilder verbose = null;
    if (logger.isTraceEnabled(LogMarker.TOMBSTONE_VERBOSE)) {
      verbose = new StringBuilder();
      verbose.append("processing tag for key ").append(getKey()).append(", stamp=")
          .append(stamp.asVersionTag()).append(", tag=").append(tag);
    }
    long stampTime = stamp.getVersionTimeStamp();
    long tagTime = tag.getVersionTimeStamp();
    if (stampTime > 0 && (tagTime > stampTime || (tagTime == stampTime
        && tag.getDistributedSystemId() >= stamp.getDistributedSystemId()))) {
      if (verbose != null) {
        verbose.append(" - allowing event");
        logger.trace(LogMarker.TOMBSTONE_VERBOSE, verbose);
      }
      // Update the stamp with event's version information.
      applyVersionTag(r, stamp, tag, originator);
      return;
    }

    if (stampTime > 0) {
      if (verbose != null) {
        verbose.append(" - disallowing event");
        logger.trace(LogMarker.TOMBSTONE_VERBOSE, verbose);
      }
      r.getCachePerfStats().incConflatedEventsCount();
      persistConflictingTag(r, tag);
      throw new ConcurrentCacheModificationException("conflicting event detected");
    }
  }
}

If I remove that code the concurrency checks behave as expected.  Tests that were added along with this change pass with the change removed.