Add load-testing CLI flags to transaction-bench#76
Draft
stablebits wants to merge 4 commits into
Draft
Conversation
Adds an optional `--num-transactions` limit that stops the benchmark after scheduling the requested number of transactions. May be combined with `--duration`; whichever limit is reached first stops the run. The generator tracks a running `txs_scheduled` count and clamps the final batch so it never overshoots the budget. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds an optional `--initial-congestion-window` (bytes) that is forwarded to tpu-client-next as `override_initial_congestion_window`. A larger window skips TCP-style slow-start at connection startup so transactions can be sent at full rate immediately. Defaults to tpu-client-next's built-in value when unset. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Each tpu-client-next instance opens one QUIC connection per leader, so the number of connections to a pinned leader equals the number of instances. Previously the only way to open N connections under one identity was to repeat --staked-identity-file N times, which is impractical for large N. Add --clients-per-identity (default 1): total instances become max(1, num staked identities) * clients-per-identity, and identities are round-robined across instances via `get(i % num_identities)` (yielding None / unstaked when no identity file is given). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
After the generator finishes there can still be transactions queued in tpu-client-next's worker channels and quinn send buffers. Tearing the schedulers down immediately drops them, which is especially visible when --num-transactions is set. Add a drain phase (--drain-seconds, default 10, 0 to disable) that waits until the aggregated successfully_sent count is stable across reporter ticks, then a short fixed tail to let quinn flush, capped by the timeout. To support this, report_aggregated_stats now also accumulates non-resetting totals (successfully_sent / congestion_events / write_error) that the drain phase observes, logs per-tick and final totals, and breaks down connection_error / write_error by cause. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds four independent enhancements to transaction-bench, aimed at driving and measuring high connection-count load tests against a single (typically pinned) leader. Each is a separate commit and works on its own.
--num-transactionsOptional cap on the total number of transactions scheduled, after which the generator stops. Can be combined with --duration — whichever limit is hit first ends the run. The generator tracks a running count and clamps the final batch so it
never overshoots the budget.
--initial-congestion-windowOverrides the QUIC initial congestion window (in bytes) passed to tpu-client-next as override_initial_congestion_window. A larger window skips TCP-style slow-start at connection startup so transactions go out at full rate immediately. Defaults
to tpu-client-next's built-in value (128 * PACKET_DATA_SIZE) when unset.
--clients-per-identityEach tpu-client-next instance opens exactly one QUIC connection per leader (WorkersCache is keyed by leader SocketAddr, one worker = one connection). So the number of connections to a pinned leader equals the number of instances, and previously the only way to get N connections under one identity was to repeat --staked-identity-file N times — impractical for large N.
This flag (default 1) sets a per-identity multiplier:
total instances = max(1, num staked identities) * clients-per-identity
Identities are round-robined across instances (get(i % num_identities)), yielding None/unstaked when no identity file is given. Example — 1000 connections under one identity:
solana-transaction-bench -u $CLUSTER read-accounts-run
--staked-identity-file $STAKED_ID --clients-per-identity 1000
... pinned-leader-tracker $TARGET
--drain-secondsPost-generation drain phase + detailed stats (--drain-seconds)
After the generator finishes, transactions can still be queued in tpu-client-next's worker channels and quinn send buffers. Tearing the schedulers down immediately drops them — most visible when --num-transactions is set. The drain phase
(default 10s, 0 to disable) waits until the aggregated successfully_sent count is stable across reporter ticks, then a short fixed tail to let quinn flush, all capped by the timeout.
To support this, report_aggregated_stats now also accumulates non-resetting totals (successfully_sent / congestion_events / write_error) that the drain phase observes, logs per-tick and final totals, and breaks down connection_error /
write_error by cause.