Firefly: send to flow destination and optional collector

[ShawnMcKee]

Summary

This PR updates firefly sending semantics to use existing configuration variables while preserving default behavior.

• pool.enable.firefly=true: continue sending fireflies to the transfer flow destination (UDP 10514).
• pool.firefly.destination unset: no extra copy is sent.
• pool.firefly.destination set: send one additional copy to that destination.

The configured pool.firefly.destination accepts:

• host (defaults to UDP 10514)
• host:port (explicit port override)

Code changes

• modules/dcache/src/main/java/org/dcache/pool/movers/TransferLifeCycle.java
  • Add optional collector destination field.
  • Keep primary flow-destination send behavior.
  • Send an additional copy to configured collector when set.
  • Avoid duplicate send when collector equals primary destination.
  • Document setFireflyDestination semantics and accepted formats.
• skel/share/defaults/pool.properties
  • Update comments for pool.firefly.destination to describe new behavior and examples.

Build validation

Executed:

• mvn -pl modules/dcache -am -DskipTests compile
• mvn -pl modules/dcache -am -DskipTests package

Result:

• BUILD SUCCESS

Runtime validation (UMFS19 + dcdum01)

Deployed updated dcache-core-11.2.1.jar on umfs19 and restarted:

• dcache@umfs19Domain_1.service active
• dcache@umfs19Domain_2.service active

Runtime config used:

• pool.enable.firefly=true
• pool.firefly.destination=firefly-collector.cern.ch

Deterministic protocol tests:

• xrootd with scitag.flow=313
• WebDAV with SciTag: 313

Capture (/tmp/firefly-collector-test.pcap) showed destinations:

• flow destination: 2001:48a8:68f7:1:192:41:231:128:10514
• collector destination: 2001:1458:301:86::100:8:10514 (firefly-collector.cern.ch)

Counts by protocol/state (aggregated across both destinations):

• xrootd start: 2
• xrootd end: 2
• https start: 2
• https end: 2

Per destination each protocol/state appeared exactly once, confirming dual-send:

• one copy to flow destination
• one copy to configured collector

[Copilot]

Pull request overview

Updates pool “firefly” UDP marker sending so the default behavior (send to transfer flow destination/UDP 10514) remains unchanged while allowing an optional second copy to be sent to a separately configured collector.

Changes:

• Add an optional “collector” destination and send firefly markers to both primary destination and collector (when configured and different).
• Keep the primary destination logic unchanged; avoid sending a duplicate when collector equals primary.
• Update pool.firefly.destination documentation to describe the new “additional collector copy” behavior and provide examples.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
modules/dcache/src/main/java/org/dcache/pool/movers/TransferLifeCycle.java	Adds optional collector destination and routes start/end markers to primary + optional collector.
skel/share/defaults/pool.properties	Updates property comments to document new pool.firefly.destination semantics and examples.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

With the new dual-send behavior, sendToMultipleDestinations may call send twice per marker, and send currently creates a new DatagramSocket each time. When a collector is configured, this doubles socket creation overhead for every start/end marker. Consider reusing a single socket per marker (send to primary and collector from the same socket) to reduce per-transfer overhead.

Suggested change

	* Send flow marker to the primary destination and optional configured collector.
	*
	* @param primaryDst Primary destination (based on flow/peer)
	* @param payload the marker
	*/
	private void sendToMultipleDestinations(InetSocketAddress primaryDst, @Nonnull String payload) {
	// Send to primary destination
	send(primaryDst, payload);
	// Send to collector if configured and different from primary
	if (fireflyCollector != null && !fireflyCollector.equals(primaryDst)) {
	send(fireflyCollector, payload);
	* Send flow marker using an existing datagram socket.
	*
	* @param socket Datagram socket to use for sending.
	* @param dst Inet address where the flow marker should be sent.
	* @param payload the marker
	* @throws IllegalStateException if flow marker ist not build.
	*/
	private void send(DatagramSocket socket, InetSocketAddress dst, @Nonnull String payload)
	throws IllegalStateException {
	byte[] data = payload.getBytes(StandardCharsets.UTF_8);
	DatagramPacket p = new DatagramPacket(data, data.length, dst);
	try {
	socket.send(p);
	} catch (IOException e) {
	LOGGER.warn("Failed to send flow marker to {}: {}", dst, e.getMessage());
	}
	}
	/**
	* Send flow marker to the primary destination and optional configured collector.
	*
	* @param primaryDst Primary destination (based on flow/peer)
	* @param payload the marker
	*/
	private void sendToMultipleDestinations(InetSocketAddress primaryDst, @Nonnull String payload) {
	try (DatagramSocket socket = new DatagramSocket()) {
	// Send to primary destination
	send(socket, primaryDst, payload);
	// Send to collector if configured and different from primary
	if (fireflyCollector != null && !fireflyCollector.equals(primaryDst)) {
	send(socket, fireflyCollector, payload);
	}
	} catch (IOException e) {
	LOGGER.warn("Failed to send flow marker: {}", e.getMessage());

[DmitryLitvintsev]

null is default initializer

[DmitryLitvintsev]

This is not needed.

[ShawnMcKee]

Thanks for the suggestions! Addressed all comments in commit 9f4b114ed6:

• fireflyCollector = null — dropped the explicit = null field initializer (default for object references).
• fireflyCollector = null; inside setter — removed the redundant reset line at the top of setFireflyDestination.
• Blank-string guard — replaced Strings.isNullOrEmpty(addr) with addr != null && !addr.isBlank() to correctly handle whitespace-only strings; also removed the now-unused Strings import.
• IPv6 Javadoc — added [ipv6-address] and [ipv6-address]:port as documented accepted formats, with a note that the bracketed form is required by HostAndPort.fromString when a port is present.
• Socket reuse — refactored send to accept a caller-supplied DatagramSocket; sendToMultipleDestinations now opens one socket and uses it for both the primary and collector sends, halving socket creation overhead when a collector is configured.

[DmitryLitvintsev]

Looks good to me. I will also wait for our colleagues in Europe to have a look as well tomorrow.

[kofemann]

@ShawnMcKee, thanks for the contribution.

[kofemann]

The PR is against 11.2. Will the update to master contain the same changes or should I cherry-pick them myself?

[ShawnMcKee]

Can you cherry-pick them? Then we can delete the branch.