[IGNITE-12094] Scan Query scheduled to be executed on wrong node

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.7.5
• Component/s: cache
• Ignite Flags: Docs Required
• Assignee: Unassigned
• Reporter: Dmitry Sterinzat
• Created: 22/Aug/19 06:47
• Updated: 22/Aug/19 06:48

Description

When Scan query on Replicated cache executes from client node and topology is unstable (for example 2 new server nodes are up at the same time) it is possible to get empty result:

// IgniteCacheProxyImpl#projection

// here we return cluster group with random cluster node 
return ctx.kernalContext().grid().cluster().forDataNodes(ctx.name()).forRandom();

 

// GridCacheQueryAdapter#executeScanQuery

// Affinity nodes snapshot.
Collection<ClusterNode> nodes = new ArrayList<>(nodes());
...
if (nodes.isEmpty()) {
    if (part != null && forceLocal)
        throw new IgniteCheckedException("No queryable nodes for partition " + part
            + " [forced local query=" + this + "]");

    return new GridEmptyCloseableIterator();
} 

// GridCacheQueryAdapter#nodes(final GridCacheContext<?, ?> cctx,
    @Nullable final ClusterGroup prj, @Nullable final Integer part)

    final AffinityTopologyVersion topVer = cctx.affinity().affinityTopologyVersion();

// This collection isn't contains randomly selected node because AffinityTopologyVersion is previous.
    Collection<ClusterNode> affNodes = CU.affinityNodes(cctx, topVer);
...
    return F.view(affNodes, new P1<ClusterNode>() {
        @Override public boolean apply(ClusterNode n) {
            return cctx.discovery().cacheAffinityNode(n, cctx.name()) &&
                (prj == null || prj.node(n.id()) != null) &&
                (part == null || owners.contains(n));
        }
    });
}

In our case nodes collection is empty, because node that was randomly selected isn't current topology version (cache isn't already started), so we get GridEmptyCloseableIterator (query event isn't executed).