contrueCT opened a new issue, #3467: URL: https://github.com/apache/hugegraph/issues/3467
## Problem In an HStore graph, REST `kout` and Gremlin neighbor queries temporarily returned HTTP 200 with empty results for vertices that had known outgoing `person_knows_person` edges. After restarting the graph/server instance, the same IDs returned the expected 1, 8, and 100 neighbors. A 61-request same-JVM sequence after restart was fully correct. The current `CachedGraphTransaction.fetchEdgeBatch()` stores an empty backend batch in the edge cache. With the default edge cache enabled and a 600-second expiry, a transient empty backend read becomes a negative cache entry. A restart clears it. The original HStore/PD event that produced the first empty backend read was not captured, so the issue should distinguish the cache defect from any underlying routing or store-read race. ## Expected behavior An empty backend read for an edge query must not permanently suppress a later successful read for the same query key. Positive edge results may continue to use the existing cache. ## Reproduction context - Latest master inspected: `cd7f1493b15858937d3684d72706d9681860bb9d`. - Deterministic unit reproduction: force the first backend query to return an empty iterator, restore the backend, and issue the same query again. Before the fix the second read returned zero edges; after the fix it returned the persisted edge. - Both REST `kout` and Gremlin vertex steps use the graph edge query path and can reach the same edge cache. ## Acceptance criteria - Empty edge batches are not retained as negative cache entries. - Positive edge cache hits, expiry handling, mutation invalidation, and query limits remain unchanged. - A regression test demonstrates recovery after a transient empty backend response. - HStore/PD routing behavior remains unchanged; any underlying transient read cause is documented separately if reproduced. -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
