rotor: clean up tile by emwang-jump · Pull Request #11583 · firedancer-io/firedancer

emwang-jump · 2026-09-17T19:36:57Z

No description provided.

github-actions · 2026-09-17T19:37:20Z

 ┌─ ⚡ PERF · 8b5dacf vs main@b12f769 ─────────────────────────────────
 │ SUITE                               BASELINE          NEW         Δ
 │ replay tps, mainnet               29,489 tps   29,630 tps  ·  +0.48%
 │ bench tps, localnet              764,124 tps  759,416 tps  ·  -0.62%
 │ snapshot load, testnet               14.74 s      14.65 s  ·  -0.56%
 │ mem total, mainnet                158.38 GiB   158.38 GiB  ·   0.00%
 │ mem total, testnet                 89.38 GiB    89.38 GiB  ·   0.00%
 │ mem total, ag mainnet             189.86 GiB   189.86 GiB  ·   0.00%
 │ mem total, ag testnet             120.86 GiB   120.86 GiB  ·   0.00%
 │ clean compile, firedancer             2.91 s       2.96 s  ·  +1.79%
 │ binary size, firedancer             56.06 MB     56.05 MB  ·  -0.03%
 ├─────────────────────────────────────────────────────────────────────
@@ 0 REGRESSIONS · 0 WARNINGS · 0 IMPROVED · 9 NOISE @@
 └─────────────────────────────────────────────────────────────────────

Copilot

🟡 Changes recommended

Block finalization and duplicate-version handling can strand repairs or corrupt replay’s live block mapping.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Cleans up the Firedancer rotor tile by consolidating repair components and separating scheduling, request generation, chaining, and statistics.

Changes:

Moves chainer code into the rotor module and simplifies block lifecycle management.
Adds dedicated scheduler, requestor, and repair-statistics components.
Replaces the rotor tile test with focused component tests.

File summaries

File	Description
`src/discof/rotor/test_schedulor.c`	Adds scheduler lifecycle and ordering tests.
`src/discof/rotor/test_requestor.c`	Adds request-generation tests.
`src/discof/rotor/test_chainer.c`	Updates chainer tests for block terminology and behavior.
`src/discof/rotor/Local.mk`	Registers new components and tests.
`src/discof/rotor/fd_schedulor.h`	Updates scheduler contract.
`src/discof/rotor/fd_schedulor.c`	Makes duplicate scheduling a no-op.
`src/discof/rotor/fd_rotor_tile_private.h`	Removes the metadata queue.
`src/discof/rotor/fd_requestor.h`	Expands request-walk behavior and chainer integration.
`src/discof/rotor/fd_requestor.c`	Implements bounded orphan filling.
`src/discof/rotor/fd_repair_stats.h`	Introduces per-slot repair statistics.
`src/discof/rotor/fd_repair_stats.c`	Implements repair-statistics storage and reporting.
`src/discof/rotor/fd_chainer.h`	Renames slot versions to blocks and removes worklists.
`src/discof/rotor/fd_chainer.c`	Relocates and simplifies chainer implementation.
`src/discof/chainer/Local.mk`	Removes the old chainer build definition.
`src/discof/chainer/fd_chainer.c`	Removes the previous chainer implementation.
`src/app/firedancer-dev/commands/rotor.c`	Updates rotor diagnostics for the new block layout.

Review details

Files reviewed: 16/18 changed files
Comments generated: 6
Review effort level: Balanced

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

+  FD_TEST( slot>chainer->root );
+  if( FD_UNLIKELY( fd_chainer_block_query( chainer, slot, &block_id ) ) ) return NULL; /* already have it */
+  return acquire_block( chainer, slot, &block_id );


+      if( FD_UNLIKELY( turbine->complete_idx!=UINT_MAX &&
+                       turbine->buffered_fec_idx==turbine->complete_idx &&
+                       fd_hash_check_zero( &turbine->block_id ) ) ) {
+        if( FD_UNLIKELY( !finalize_block_id( chainer, turbine ) ) ) FD_LOG_WARNING(( "failed to finalize block_id for slot %lu, parent_slot %lu parent_bid is zero %d", slot, turbine->parent_slot, fd_hash_check_zero( &turbine->parent_block_id ) ));
+      }


+  fd_chainer_block_t * parent_block = fd_chainer_block_query( chainer, parent_slot, parent_block_id );
+  if( FD_UNLIKELY( !parent_block ) ) {
+    parent_block = acquire_block( chainer, parent_slot, parent_block_id );
+    created      = parent_block;
+  }


+  fd_repair_stats_slot_t * r = &self->slots[ slot % self->slot_max ];
+  if( FD_LIKELY( r->slot==slot ) ) return r;
+  *r = (fd_repair_stats_slot_t){ .slot = slot, .first_shred_ts = now };
+  return r;
+}
+
+void
+fd_repair_stats_slot_start( fd_repair_stats_t * self,
+                            ulong               slot,
+                            long                now ) {
+  record_start( self, slot, now );
+}
+
+void
+fd_repair_stats_shred_received( fd_repair_stats_t * self,
+                                ulong               slot,
+                                int                 is_data,
+                                uint                shred_src,
+                                long                now ) {
+  fd_repair_stats_slot_t * r = record_start( self, slot, now );


@@ -1,6 +1,11 @@
+$(call add-hdrs,fd_rotor_tile.h fd_chainer.h fd_schedulor.h fd_requestor.h fd_repair_stats.h)
+$(call add-objs,fd_chainer fd_schedulor fd_requestor fd_repair_stats,fd_discof)


Copilot

🟡 Changes recommended

Inflight key classification, scheduler rekeying, rooted-parent handling, and repair-statistics behavior have unresolved correctness issues.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (4)

Previously missed (1) — in code that hasn't changed since the last review.

src/discof/rotor/fd_repair_stats.c:149

This logs every completion passed in, but records are shared by all versions of a slot and contain no completed/reported state. Since the new chainer deliberately allows multiple versions to complete and deliver, the same slot can emit multiple “complete” lines with different durations, contrary to the documented “first to complete stops it” behavior. Persist the first completion (or a reported flag) in the slot record and ignore later completions.

src/discof/rotor/fd_chainer.c:460

Finalizing the turbine block changes its scheduler identity from {slot, ZERO} to {slot, block_id}, but this function does not expose that rename and after_fec_complete only schedules newly created blocks. The existing zero-key task therefore becomes stale and is dropped; a completed turbine block with a missing parent will never be walked under its finalized ID to continue orphan repair. Report the rekey to the tile (or update the scheduler directly) so the old task is removed and the finalized key is queued.

      if( FD_UNLIKELY( turbine->complete_idx!=UINT_MAX &&
                       turbine->buffered_fec_idx==turbine->complete_idx &&
                       fd_hash_check_zero( &turbine->block_id ) ) ) {
        if( FD_UNLIKELY( !finalize_block_id( chainer, turbine ) ) ) FD_LOG_WARNING(( "failed to finalize block_id for slot %lu, parent_slot %lu parent_bid is zero %d", slot, turbine->parent_slot, fd_hash_check_zero( &turbine->parent_block_id ) ));

src/discof/rotor/fd_repair_stats.c:110

A delayed older slot unconditionally overwrites a newer slot that occupies the same modulo bucket, even though the API promises replacement only when a newer slot lands. Since the rotor can retain more than slot_max slots and shreds arrive out of order, this can erase the active newer slot's timing and counters. Reject stale slot<r->slot updates (and make callers handle the ignored record) instead of replacing it.

  fd_repair_stats_slot_t * r = &self->slots[ slot % self->slot_max ];
  if( FD_LIKELY( r->slot==slot ) ) return r;
  *r = (fd_repair_stats_slot_t){ .slot = slot, .first_shred_ts = now };

src/discof/rotor/fd_chainer.c:530

When the verified parent is absent but already at or below the current root, creating it leaves a block below the root (or a noncanonical root sibling). This contradicts fd_chainer_verify's invariant and restores state for a dead fork; the previous implementation explicitly returned without allocating in this case. Return NULL when parent_slot<=chainer->root before acquiring the parent.

  fd_chainer_block_t * parent_block = fd_chainer_block_query( chainer, parent_slot, parent_block_id );
  if( FD_UNLIKELY( !parent_block ) ) {
    parent_block = acquire_block( chainer, parent_slot, parent_block_id );
    created      = parent_block;

Files reviewed: 19/22 changed files
Comments generated: 1
Review effort level: Balanced

                           long                now ) {
  fd_inflight_t * req = inflight_acquire( table );
-  fd_inflight_key_init( &req->key, FD_REPAIR_KIND_SHRED, slot, shred_idx, nonce, fec_root );
+  fd_inflight_key_init( &req->key, kind, slot, shred_idx, nonce, fec_root );


Copilot

🟡 Changes recommended

Inflight keying, turbine scheduler rekeying, rooted-parent handling, and completion statistics contain correctness regressions.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (3)

src/discof/repair/fd_inflight.c:123

kind can now be AG_REPAIR_KIND_SHRED_FOR_BLOCK_ID, but fd_inflight_key_is_shred still returns true only for FD_REPAIR_KIND_SHRED. These requests therefore take the metadata nonce-only equality/hash path, so slot, shred index, FEC root, and even request kind are ignored; a nonce collision can remove and credit the wrong inflight request. Update the key classification/equality so both shred kinds use the full shred key (including a consistent treatment of kind).

  fd_inflight_key_init( &req->key, kind, slot, shred_idx, nonce, fec_root );

src/discof/rotor/fd_chainer.c:460

Finalizing a turbine block changes its scheduler identity from {slot, ZERO} to {slot, block_id}, but this mutation is not reported to the tile. For the normal case where the block was created by earlier shreds, the tile only has the old queued key; that stale check is dropped after finalization and no check is queued for the new key. If the completed block's parent is still absent, ancestry retries stop and the block can remain permanently undelivered. Rekey the scheduler entry when finalization succeeds or expose the rename so the tile can do so.

      if( FD_UNLIKELY( turbine->complete_idx!=UINT_MAX &&
                       turbine->buffered_fec_idx==turbine->complete_idx &&
                       fd_hash_check_zero( &turbine->block_id ) ) ) {
        if( FD_UNLIKELY( !finalize_block_id( chainer, turbine ) ) ) FD_LOG_WARNING(( "failed to finalize block_id for slot %lu, parent_slot %lu parent_bid is zero %d", slot, turbine->parent_slot, fd_hash_check_zero( &turbine->parent_block_id ) ));

src/discof/rotor/fd_chainer.c:530

When the named parent is no longer retained because parent_slot<=root, creating it reintroduces a pruned block (and for parent_slot<root, immediately violates fd_chainer_verify's no-block-below-root invariant). At the root, each dead-fork parent ID can also consume another version until the per-slot limit aborts. Treat an absent parent at or below the root as already settled instead of allocating it, as the previous implementation did.

  fd_chainer_block_t * parent_block = fd_chainer_block_query( chainer, parent_slot, parent_block_id );
  if( FD_UNLIKELY( !parent_block ) ) {
    parent_block = acquire_block( chainer, parent_slot, parent_block_id );
    created      = parent_block;

Files reviewed: 20/23 changed files
Comments generated: 1
Review effort level: Balanced

+  fd_repair_stats_slot_t const * r = &self->slots[ slot % self->slot_max ];
+  if( FD_UNLIKELY( r->slot!=slot ) ) return;
+  FD_LOG_NOTICE(( "slot %lu complete in %ld ms: turbine %u repair %u recovered %u code %u",
+                  slot, ( complete_ts - r->first_shred_ts )/1000000L,
+                  r->turbine_cnt, r->repair_cnt, r->recovered_cnt, r->code_cnt ));


Copilot

🟡 Changes recommended

Inflight keying, turbine finalization scheduling, replay identity handling, and repair statistics contain correctness regressions.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (4)

Previously missed (1) — in code that hasn't changed since the last review.

src/discof/rotor/fd_repair_stats.c:129

This counts every shred notification, including duplicates: the shred tile explicitly forwards FD_FEC_RESOLVER_SHRED_DUPLICATE (fd_shred_tile.c:1096-1100). The legacy forest statistics this sidecar replaces increment source counts only for newly seen shred indices (fd_forest.c:1367-1370). Repeated packets will therefore inflate turbine/repair/recovered and coding counts; pass duplicate/newness information or deduplicate by shred index before incrementing.

src/discof/rotor/fd_chainer.c:625

Creating the verified sibling without abandoning or otherwise deduplicating the incomplete turbine sibling allows both delivery streams to complete with the same {slot, block_id}. Replay still explicitly relies on the old one-stream guarantee (fd_replay_tile.c:3871-3885): turbine finalization removes any existing entry for that key and inserts its own, even when the existing verified bank is live. Restore single-stream behavior or update replay to merge/reject the duplicate live bank before enabling both versions.

  FD_TEST( slot>chainer->root );
  if( FD_UNLIKELY( fd_chainer_block_query( chainer, slot, &block_id ) ) ) return NULL; /* already have it */
  return acquire_block( chainer, slot, &block_id );

src/discof/rotor/fd_chainer.c:460

Finalization changes the turbine block's identity from {slot, 0} to {slot, block_id}, but the tile's queued schedulor task remains keyed by {slot, 0}. The next check is consequently treated as stale and dropped, with no task inserted for the finalized key; if this completed block still lacks its parent, ancestry retries stop and it may never connect or deliver. Re-key the schedulor task when finalization succeeds.

      if( FD_UNLIKELY( turbine->complete_idx!=UINT_MAX &&
                       turbine->buffered_fec_idx==turbine->complete_idx &&
                       fd_hash_check_zero( &turbine->block_id ) ) ) {
        if( FD_UNLIKELY( !finalize_block_id( chainer, turbine ) ) ) FD_LOG_WARNING(( "failed to finalize block_id for slot %lu, parent_slot %lu parent_bid is zero %d", slot, turbine->parent_slot, fd_hash_check_zero( &turbine->parent_block_id ) ));

src/discof/rotor/fd_repair_stats.c:110

A different slot always overwrites this bucket, even when it is older than the resident slot. Shreds can arrive out of order, so a delayed slot that is still above root can erase a newer colliding record and its subsequent packets can corrupt the newer slot's statistics, contrary to the documented newer-slot-only eviction policy. Preserve the resident record when the incoming slot is older (and have callers ignore that update).

  fd_repair_stats_slot_t * r = &self->slots[ slot % self->slot_max ];
  if( FD_LIKELY( r->slot==slot ) ) return r;
  *r = (fd_repair_stats_slot_t){ .slot = slot, .first_shred_ts = now };

Files reviewed: 20/23 changed files
Comments generated: 1
Review effort level: Balanced

  uint  idx;        /* shred idx (shred kinds) or fec_set_idx (AG_REPAIR_KIND_FEC_ROOT) */
  uint  nonce;      /* rnonce or counter nonce (metadata) */
-  uint  kind;       /* FD_REPAIR_KIND_SHRED for every shred request, else AG_REPAIR_KIND_{PARENT_FEC_COUNT,FEC_ROOT} */
+  uint  kind;       /* FD_REPAIR_KIND_SHRED, AG_REPAIR_KIND_SHRED_FOR_BLOCK_ID, AG_REPAIR_KIND_PARENT_FEC_COUNT or AG_REPAIR_KIND_FEC_ROOT */


Copilot

🔵 Needs a closer look

Inflight key classification, duplicate replay streams, and schedulor key transitions can misroute responses or strand blocks.

Review details

Suppressed comments (9)

Previously missed (4) — in code that hasn't changed since the last review.

src/discof/rotor/fd_chainer.h:393

This summary contradicts the documented NULL/unknown-ID behavior on lines 383-384 and the implementation, which keeps all versions of the new root when no canonical block is found. Preserve that exception here so callers do not infer that all siblings are always removed.
src/discof/rotor/fd_chainer.h:187
The magic constant name is incorrectly lowercased.

This issue also appears on line 198 of the same file.
src/discof/rotor/fd_chainer.c:6

This reusable Firedancer component introduces buffered stdio streaming through printf/fflush. That bypasses the repository's fd_io path and can add libc syscall behavior incompatible with strict tile seccomp profiles. Convert fd_chainer_print to the existing fd_io output helpers and remove this include.
src/discof/rotor/fd_chainer.c:72
fd_chainer_join(NULL) dereferences chainer while reading magic, unlike the other join APIs in this PR. This turns invalid input into a null-pointer crash instead of returning NULL.

src/discof/repair/fd_inflight.h:52

AG_REPAIR_KIND_SHRED_FOR_BLOCK_ID is still classified as metadata by fd_inflight_key_is_shred (line 86), which only recognizes FD_REPAIR_KIND_SHRED. Consequently these new records compare and hash by nonce alone, so a same-nonce metadata request can be consumed by a shred response (and vice versa), while slot/index/FEC-root keying is ignored. Classify both shred kinds as shred keys.

  uint  kind;       /* FD_REPAIR_KIND_SHRED, AG_REPAIR_KIND_SHRED_FOR_BLOCK_ID, AG_REPAIR_KIND_PARENT_FEC_COUNT or AG_REPAIR_KIND_FEC_ROOT */

src/discof/rotor/fd_chainer.c:625

Keeping an unfinished turbine block live when the verified version is inserted violates replay's one-delivery-stream contract. If the votor ID is the ID the turbine block later computes (the normal non-equivocation case), replay already has {slot, block_id} from the verified stream; the turbine slot-complete FEC is then skipped or rekeys {slot,0} over that live entry (fd_replay_tile.c:3865-3885). Restore abandonment/deduplication so only one stream can complete for a block ID.

fd_chainer_verified_block_insert( fd_chainer_t * chainer,
                                  ulong          slot,
                                  fd_hash_t      block_id ) {
  FD_TEST( slot>chainer->root );
  if( FD_UNLIKELY( fd_chainer_block_query( chainer, slot, &block_id ) ) ) return NULL; /* already have it */
  return acquire_block( chainer, slot, &block_id );

src/discof/rotor/fd_chainer.c:460

Finalizing a turbine block mutates its schedulor identity from {slot,0} to {slot,block_id}, but the tile is not notified to move the queued check. For a complete block whose known parent is absent, the old-key check later finds nothing and is dropped, so no further Orphan request is made and the block can remain disconnected indefinitely. Return/emit the key transition and atomically remove the zero-key check and queue the finalized key.

      if( FD_UNLIKELY( turbine->complete_idx!=UINT_MAX &&
                       turbine->buffered_fec_idx==turbine->complete_idx &&
                       fd_hash_check_zero( &turbine->block_id ) ) ) {
        if( FD_UNLIKELY( !finalize_block_id( chainer, turbine ) ) ) FD_LOG_WARNING(( "failed to finalize block_id for slot %lu, parent_slot %lu parent_bid is zero %d", slot, turbine->parent_slot, fd_hash_check_zero( &turbine->parent_block_id ) ));

src/discof/rotor/fd_requestor.c:278

some_parent_exists treats any version at the parent slot as the required parent. If only an equivocating sibling exists, this disables the orphan fill bound and can emit every missing shred before the Orphan request, contradicting the exact {parent_slot,parent_block_id} ancestry check and the API's documented bound. Base parent_seen only on parent_orphaned.
src/discof/rotor/fd_chainer.h:200
The constant name is incorrectly lowercased in this public API documentation.

Files reviewed: 24/29 changed files
Comments generated: 0 new
Review effort level: Balanced

Copilot

🟡 Changes recommended

Inflight key classification, duplicate block streams, and scheduler rekeying introduce correctness and reliability failures.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (4)

src/discof/repair/fd_inflight.c:123

AG_REPAIR_KIND_SHRED_FOR_BLOCK_ID is still classified as metadata by fd_inflight_key_is_shred, which only recognizes FD_REPAIR_KIND_SHRED. Consequently these records hash and compare by nonce alone: a response with the echoed nonce matches even when its slot, index, or FEC root differs, and a nonce collision can consume an unrelated metadata request. Treat both shred kinds as shred keys and include kind in shred-key equality so it remains consistent with hashing the full key.

  fd_inflight_key_init( &req->key, kind, slot, shred_idx, nonce, fec_root );

src/discof/rotor/fd_chainer.c:625

Keeping both versions live is unsafe when the votor ID equals the eventual turbine ID (the normal no-equivocation race). Both streams can reach replay as {slot, block_id}; replay still assumes the turbine stream was abandoned and removes any existing entry under that key as “stale” (fd_replay_tile.c:3871-3885), which can detach a live verified bank. Either retain the turbine-abandon/coalescing behavior or update replay to safely merge this case before allowing both versions to deliver.

  FD_TEST( slot>chainer->root );
  if( FD_UNLIKELY( fd_chainer_block_query( chainer, slot, &block_id ) ) ) return NULL; /* already have it */
  return acquire_block( chainer, slot, &block_id );

src/discof/rotor/fd_requestor.c:232

This increment is unreachable because the turbine branch immediately above always returns. The block_id_only suppression case is already handled before the loop, so leaving this line suggests a control path that cannot occur.
src/discof/rotor/fd_chainer.c:460
Finalizing a turbine block changes its scheduler identity from {slot, 0} to {slot, block_id}, but this API does not report that rekey and the rotor only schedules newly created blocks. The queued zero-ID check therefore becomes stale; if the completed block's parent is absent, it stops issuing Orphan retries and cannot connect or deliver. Report the rekey to the tile and migrate the queued check (remove the zero key and insert the finalized key).

      if( FD_UNLIKELY( turbine->complete_idx!=UINT_MAX &&
                       turbine->buffered_fec_idx==turbine->complete_idx &&
                       fd_hash_check_zero( &turbine->block_id ) ) ) {
        if( FD_UNLIKELY( !finalize_block_id( chainer, turbine ) ) ) FD_LOG_WARNING(( "failed to finalize block_id for slot %lu, parent_slot %lu parent_bid is zero %d", slot, turbine->parent_slot, fd_hash_check_zero( &turbine->parent_block_id ) ));

Files reviewed: 24/29 changed files
Comments generated: 1
Review effort level: Balanced

-    <counter name="PktTx" summary="Network packets sent, including reqs, pings, pongs, etc" />
-    <counter name="RequestTx" enum="RepairSentRequestType" summary="Client messages sent, by type" />
+    <counter name="PktTx" summary="Network packets sent, including requests, pings and pongs" />
+    <counter name="RequestTx" enum="RepairSentRequestType" summary="Repair requests sent, by type" />


Copilot

Copilot review overview

🔵 Needs a closer look

Inflight key collisions and conflicting turbine/verified block lifecycles can misroute responses or corrupt replay state.

Review effort: Balanced
Findings: 3 · 4 · 2

Open (9)

AG_REPAIR_KIND_SHRED_FOR_BLOCK_ID is now passed into the inflight key, but…
kind can now be AG_REPAIR_KIND_SHRED_FOR_BLOCK_ID, but fd_inflight_key_is_shred in…
Keeping an in-flight turbine block alive when a verified block for the same slot is inserted allows…
This does not implement the documented “first version to complete stops it” behavior.…
This overwrites the bucket for any different slot, although the API promises that only a newer…
When the response names a parent at or below the root but that exact block ID is absent, this…
Finalizing the turbine block mutates its scheduler identity from {slot, 0} to {slot, block_id},…
This metric also has a PONG enum value, and the rotor increments it when sending pong responses,…
The new stateful fd_repair_stats module is linked without any unit-test target, unlike the…

Previously missed (2)

In code that hasn't changed since last review

Correct unmatched-request gauge description

src/disco/metrics/metrics.xml:1436

This description now implies that every unmatched request is included, but the rotor does not insert HighestShred or Orphan requests into ctx->rtt (fd_rotor_tile.c:312-320). The gauge still excludes those request types, so dashboards will undercount relative to the documented meaning.

Align repair response counter with counted request types

src/disco/metrics/metrics.xml:1439

The implementation increments this counter only when fd_rnonce_ss_normal_repair(nonce) is true (fd_rotor_tile.c:717-720), so data returned by HighestShred or Orphan requests is excluded even though it is also a repair response. Narrow the summary to the requests actually counted.

Copilot AI balanced review requested due to automatic review settings September 17, 2026 19:36

Copilot started reviewing on behalf of emwang-jump September 17, 2026 19:37 View session

Copilot AI reviewed Sep 17, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings September 17, 2026 23:52

Copilot started reviewing on behalf of emwang-jump September 17, 2026 23:53 View session

Copilot AI reviewed Sep 17, 2026

View reviewed changes

Comment thread src/discof/repair/fd_inflight.c

long now ) {

fd_inflight_t * req = inflight_acquire( table );

fd_inflight_key_init( &req->key, FD_REPAIR_KIND_SHRED, slot, shred_idx, nonce, fec_root );

fd_inflight_key_init( &req->key, kind, slot, shred_idx, nonce, fec_root );

Copilot AI review requested due to automatic review settings September 18, 2026 14:59

Copilot started reviewing on behalf of emwang-jump September 18, 2026 15:00 View session

Copilot AI reviewed Sep 18, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings September 18, 2026 15:16

Copilot started reviewing on behalf of emwang-jump September 18, 2026 15:16 View session

Copilot AI reviewed Sep 18, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings September 18, 2026 19:32

Copilot started reviewing on behalf of emwang-jump September 18, 2026 19:32 View session

emwang-jump marked this pull request as ready for review September 18, 2026 19:34

Copilot AI reviewed Sep 18, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings September 18, 2026 19:56

emwang-jump force-pushed the emwang/rotor2 branch from 50285fb to 141dd43 Compare September 18, 2026 19:56

Copilot started reviewing on behalf of emwang-jump September 18, 2026 19:57 View session

Copilot AI reviewed Sep 18, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings September 20, 2026 23:45

Copilot started reviewing on behalf of emwang-jump September 20, 2026 23:46 View session

Copilot AI reviewed Sep 20, 2026

View reviewed changes

rotor: clean up tile

f332724

emwang-jump force-pushed the emwang/rotor2 branch from 4612ceb to f332724 Compare September 21, 2026 00:26

Navigation Menu

rotor: clean up tile - #11583

rotor: clean up tile#11583
emwang-jump wants to merge 1 commit into
mainfrom
emwang/rotor2

emwang-jump commented Sep 17, 2026

Uh oh!

github-actions Bot commented Sep 17, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		@@ -1,6 +1,11 @@
		$(call add-hdrs,fd_rotor_tile.h fd_chainer.h fd_schedulor.h fd_requestor.h fd_repair_stats.h)
		$(call add-objs,fd_chainer fd_schedulor fd_requestor fd_repair_stats,fd_discof)

Conversation

emwang-jump commented Sep 17, 2026

Uh oh!

github-actions Bot commented Sep 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Suppressed comments (4)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Suppressed comments (3)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Suppressed comments (4)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🔵 Needs a closer look

Suppressed comments (9)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Suppressed comments (4)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🔵 Needs a closer look

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions Bot commented Sep 17, 2026 •

edited

Loading