apache/hadoop · error · InvalidRequestException

there is no shared memory segment registered with shmId {}

Error message

there is no shared memory segment registered with shmId {}

What it means

ShortCircuitRegistry.registerSlot() maps a client's shared-memory slot (SlotId = shmId + slotIdx) to an ExtendedBlockId. It first looks up the ShmId in its segments map; an unknown shmId means the DataNode has no registered short-circuit shared memory segment with that id, so it throws InvalidRequestException, which is serialized back to the DFS client over the domain socket / client-datanode protocol.

Source

Thrown at hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/ShortCircuitRegistry.java:352

    if (LOG.isTraceEnabled()) {
      LOG.trace("createNewMemorySegment: created " + info.shmId);
    }
    return info;
  }
  
  public synchronized void registerSlot(ExtendedBlockId blockId, SlotId slotId,
      boolean isCached) throws InvalidRequestException {
    if (!enabled) {
      if (LOG.isTraceEnabled()) {
        LOG.trace(this + " can't register a slot because the " +
            "ShortCircuitRegistry is not enabled.");
      }
      throw new UnsupportedOperationException();
    }
    ShmId shmId = slotId.getShmId();
    RegisteredShm shm = segments.get(shmId);
    if (shm == null) {
      throw new InvalidRequestException("there is no shared memory segment " +
          "registered with shmId " + shmId);
    }
    Slot slot = shm.registerSlot(slotId.getSlotIdx(), blockId);
    if (isCached) {
      slot.makeAnchorable();
    } else {
      slot.makeUnanchorable();
    }
    boolean added = slots.put(blockId, slot);
    Preconditions.checkState(added);
    if (LOG.isTraceEnabled()) {
      LOG.trace(this + ": registered " + blockId + " with slot " +
        slotId + " (isCached=" + isCached + ")");
    }
  }
  
  public synchronized void unregisterSlot(SlotId slotId)
      throws InvalidRequestException {

View on GitHub (pinned to 2add963021)

Solutions

1. Client side: discard the stale shared-memory segment on InvalidRequestException and re-establish it (new requestShortCircuitShm) before retrying the short-circuit read
2. Falling back to normal (non-short-circuit) reads after this error is safe and automatic in most client versions - keep the fallback enabled
3. Operator side: if this floods logs after a DN restart, that is expected churn; ensure clients eventually refresh rather than pinning dead segments

// client side: only use SlotIds from segments you successfully created
if (shm == null || !shm.isValid()) { shm = requestShortCircuitShm(dn); } // re-establish before registering slots

catch (InvalidRequestException e) { // from requestShortCircuitFds with a SlotId
  if (e.getMessage().contains("no shared memory segment")) {
    discardShmAndReestablish(); // drop stale shmId, requestShortCircuitShm again, then retry once
  } else throw e;
}

When it happens

Trigger: A client sends a slot registration (block request with a SlotId) whose shmId was never established via a successful requestShortCircuitShm handshake - e.g., the segment was already torn down (DN restart, registry shutdown/re-enable, idle eviction) and the client retries with the stale shmId, or a fabricated/duplicated slot request.

Common situations: Client holds a cached ShortCircuitSharedMemorySegment across a DataNode restart or ShortCircuitRegistry flush; races between 'shutdown()' (which clears segments) and in-flight slot requests; tests hammering short-circuit reads (dfs.client.read.shortcircuit=true with dfs.domain.socket.path configured) against a restarted DN.

Related errors

• failed to allocate new BlockReader at position {}
• the datanode {} failed to pass a file descriptor (might have
• {this}: slot {slotIdx} does not exist.
• {this}: invalid negative slot index {slotIdx}
• {this}: invalid slot index {slotIdx}

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/c4ea91ea82c37932. Report an issue: GitHub.