apollo_network_benchmark: add ReveresedSqmrSender struct and implementation

[sirandreww-starkware]

Note

Medium Risk
Adds new stateful async sending behavior for the reversed-sqmr benchmark protocol, which could change stress-test traffic patterns and introduce subtle timing issues (timeouts/query replacement), but is confined to benchmarking code.

Overview
Adds a new MessageSender::ReveresedSqmr variant backed by ReveresedSqmrSender, which tracks the latest incoming SQMR server query and replies to it with outgoing broadcast messages.

send_message now supports this mode by opportunistically polling for new queries (short timeout), replacing the active query, and sending responses with improved info/trace/error logging; protocol channel registration for reversed-sqmr remains todo!().

^{Written by Cursor Bugbot for commit 4541183. This will update automatically on new commits. Configure here.}

[reviewable-StarkWare]

This change is 

[sirandreww-starkware]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• apollo_network_benchmark: implement RoundRobin broadcasting logic #11560 : 3 dependent PRs (#11561 , #12722 , #13288 )
• apollo_network_benchmark: add seconds_since_epoch helper function #11559
• apollo_network_benchmark: add RoundRobin mode to enum and round_duration_seconds arg #11558
• apollo_network_benchmark: add message index detection mechanism #11557
• apollo_network_benchmark: add MessageIndexTracker struct #11556
• apollo_network_benchmark: add ReveresedSqmr receiver and registration #11555
• apollo_network_benchmark: add ReveresedSqmrSender struct and implementation #11554 👈 (View in Graphite)
• apollo_network_benchmark: add ReveresedSqmr variant to NetworkProtocol enum #11553
• apollo_network_benchmark: add Sqmr variant to MessageReceiver and registration #11552
• apollo_network_benchmark: add Sqmr variant to MessageSender with send logic #11551
• apollo_network_benchmark: add Sqmr variant to NetworkProtocol enum #11550
• apollo_network_benchmark: add SQMR protocol constant #11549
• apollo_network_benchmark: added ping latency metrics #11548
• apollo_network_benchmark: added network event metrics #11547
• apollo_network_benchmark: added SQMR session metrics #11546
• apollo_network_benchmark: added broadcast topic metrics #11545
• apollo_network_benchmark: added network metrics to network manager #11544
• apollo_network_benchmark: added broadcast metrics tracking #11543
• apollo_network_benchmark: added conditional broadcasting based on mode #11542
• apollo_network_benchmark: added broadcasting mode selection #11541
• apollo_network_benchmark: added message broadcasting task #10734 : 1 other dependent PR (#8303 )
• apollo_network_benchmark: added register metrics #10660 : 1 other dependent PR (#10661 )
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

There hasn't been any activity on this pull request recently, and in order to prioritize active work, it has been marked as stale.
This PR will be closed and locked in 7 days if no further activity occurs.
Thank you for your contributions!

[cursor]

Only one query stored despite broadcast-to-all intent

Medium Severity

collect_new_queries replaces active_query each iteration, so only the most recent query is retained and all prior queries are silently dropped. The comment on line 143 says "broadcast the message to all active queries," and the method is named broadcast_to_queries (plural), but the struct only holds a single Option — meaning in a multi-receiver benchmark scenario, only the last peer to connect actually receives data. If broadcasting to all connected peers is intended, active_query needs to be a collection.

Additional Locations (1)

• crates/apollo_network_benchmark/src/bin/broadcast_network_stress_test_node/protocol.rs#L139-L145

 

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

[cursor]

1ms timeout per send inflates benchmark latency metrics

Medium Severity

collect_new_queries uses a tokio::time::timeout of 1ms in its while loop. When no queries are pending (the common steady-state case), the final iteration always blocks for 1ms before the timeout fails and the loop exits. Since collect_new_queries is called on every send_message invocation, this adds at least 1ms of artificial latency to every send. In a stress test benchmark, this caps throughput at ~1000 msg/s and inflates the BROADCAST_MESSAGE_SEND_DELAY_SECONDS metric recorded by the caller, producing misleading benchmark results.

 

[cursor]

Silent message drop when no active query exists

Medium Severity

broadcast_to_queries silently discards the message when active_query is None — no warning or error is logged. The caller in handlers.rs unconditionally increments BROADCAST_MESSAGE_COUNT and records send delay after calling send_message, so dropped messages are counted as successfully sent. This produces inaccurate benchmark metrics. The other protocol variants either panic or log errors on failure, making this inconsistency especially easy to miss during debugging.