firedancer-dev: repair profiler light

Author: jvarela-jump

src/app/firedancer-dev/Local.mk

@@ -14,6 +14,7 @@ $(call add-objs,commands/dev,fd_firedancer_dev)
 $(call add-objs,commands/sim,fd_firedancer_dev)
 $(call add-objs,commands/backtest,fd_firedancer_dev)
 $(call add-objs,commands/snapshot_load,fd_firedancer_dev)
+$(call add-objs,commands/repair,fd_firedancer_dev)
 
 $(call make-bin,firedancer-dev,main,fd_firedancer_dev fd_firedancer fddev_shared fdctl_shared fdctl_platform fd_discof fd_disco fd_choreo fd_flamenco fd_funk fd_quic fd_tls fd_reedsol fd_waltz fd_tango fd_ballet fd_util firedancer_version,$(SECP256K1_LIBS) $(ROCKSDB_LIBS) $(OPENSSL_LIBS))
 

src/app/firedancer-dev/commands/repair.c

@@ -180,6 +180,7 @@ extern action_t fd_action_gossip;
 extern action_t fd_action_sim;
 extern action_t fd_action_backtest;
 extern action_t fd_action_snapshot_load;
+extern action_t fd_action_repair;
 
 action_t * ACTIONS[] = {
   &fd_action_run,
@@ -208,6 +209,7 @@ action_t * ACTIONS[] = {
   &fd_action_sim,
   &fd_action_backtest,
   &fd_action_snapshot_load,
+  &fd_action_repair,
   NULL,
 };
 

src/app/firedancer-dev/main.c

@@ -789,7 +789,7 @@ fd_topo_initialize( config_t * config ) {
     fd_topob_wksp( topo, "repair_scap" );
     fd_topob_wksp( topo, "replay_scap" );
 
-    fd_topob_tile( topo, "scap", "scap", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0 );
+    fd_topo_tile_t * scap_tile = fd_topob_tile( topo, "scap", "scap", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0 );
 
     fd_topob_link( topo, "repair_scap", "repair_scap", 128UL, FD_SLICE_MAX_WITH_HEADERS, 1UL );
     fd_topob_link( topo, "replay_scap", "replay_scap", 128UL, sizeof(fd_hash_t)+sizeof(ulong), 1UL );
@@ -809,6 +809,9 @@ fd_topo_initialize( config_t * config ) {
 
     fd_topob_tile_out( topo, "repair", 0UL, "repair_scap", 0UL );
     fd_topob_tile_out( topo, "replay", 0UL, "replay_scap", 0UL );
+
+    fd_topob_tile_uses( topo, scap_tile, root_slot_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
+    /* No default fd_topob_tile_in connection to stake_out */
   }
 
   fd_topob_wksp( topo, "replay_notif" );
@@ -1105,6 +1108,8 @@ fd_topo_configure_tile( fd_topo_tile_t * tile,
       tile->shredcap.repair_intake_listen_port = config->tiles.repair.repair_intake_listen_port;
       strncpy( tile->shredcap.folder_path, config->tiles.shredcap.folder_path, sizeof(tile->shredcap.folder_path) );
       tile->shredcap.write_buffer_size = config->tiles.shredcap.write_buffer_size;
+      tile->shredcap.enable_publish_stake_weights = 0; /* this is not part of the config */
+      strncpy( tile->shredcap.manifest_path, "", PATH_MAX ); /* this is not part of the config */
     } else {
       return 0;
     }

src/app/firedancer/topology.c

@@ -67,6 +67,10 @@ union fdctl_args {
     char name[ 13UL ];
   } flame;
 
+  struct {
+    char manifest_path[ 256UL ];
+  } repair;
+
   struct {
     char    affinity[ AFFINITY_SZ ];
     uint    tpu_ip;

src/app/shared/fd_action.h

@@ -462,6 +462,8 @@ struct fd_topo_tile {
       char   folder_path[ PATH_MAX ];
       ushort repair_intake_listen_port;
       ulong   write_buffer_size; /* Size of the write buffer for the capture tile */
+      int    enable_publish_stake_weights;
+      char   manifest_path[ PATH_MAX ];
 
       /* Set internally by the capture tile */
       int shreds_fd;

src/disco/topo/fd_topo.h

@@ -842,6 +842,7 @@ after_frag( fd_repair_tile_ctx_t * ctx,
     fd_fseq_update( ctx->turbine_slot, fd_ulong_max( shred->slot, fd_fseq_query( ctx->turbine_slot ) ) );
     if( FD_UNLIKELY( shred->slot <= fd_forest_root_slot( ctx->forest ) ) ) return; /* shred too old */
 
+    /* TODO add automated caught-up test */
 
     /* Insert the shred sig (shared by all shred members in the FEC set)
        into the map. */
@@ -1075,6 +1076,7 @@ during_housekeeping( fd_repair_tile_ctx_t * ctx ) {
     return;
   }
 }
+
 static void
 privileged_init( fd_topo_t *      topo,
                  fd_topo_tile_t * tile ) {
@@ -1196,7 +1198,6 @@ unprivileged_init( fd_topo_t *      topo,
 
   /* Scratch mem setup */
 
-  ctx->blockstore = &ctx->blockstore_ljoin;
   ctx->repair     = FD_SCRATCH_ALLOC_APPEND( l, fd_repair_align(), fd_repair_footprint() );
   ctx->forest = FD_SCRATCH_ALLOC_APPEND( l, fd_forest_align(), fd_forest_footprint( tile->repair.slot_max ) );
   ctx->fec_sigs = FD_SCRATCH_ALLOC_APPEND( l, fd_fec_sig_align(), fd_fec_sig_footprint( 20 ) );
@@ -1237,17 +1238,20 @@ unprivileged_init( fd_topo_t *      topo,
   }
 
   /* Blockstore setup */
+  ctx->blockstore = NULL;
   ulong blockstore_obj_id = fd_pod_queryf_ulong( topo->props, ULONG_MAX, "blockstore" );
-  FD_TEST( blockstore_obj_id!=ULONG_MAX );
-  ctx->blockstore_wksp = topo->workspaces[ topo->objs[ blockstore_obj_id ].wksp_id ].wksp;
-
-  if( ctx->blockstore_wksp==NULL ) {
-    FD_LOG_ERR(( "no blocktore workspace" ));
+  if( FD_UNLIKELY( blockstore_obj_id==ULONG_MAX ) ) {
+    FD_LOG_WARNING(( "no blockstore_obj_id" ));
+  } else {
+    ctx->blockstore_wksp = topo->workspaces[ topo->objs[ blockstore_obj_id ].wksp_id ].wksp;
+    if( FD_UNLIKELY( ctx->blockstore_wksp==NULL ) ) {
+      FD_LOG_WARNING(( "no blocktore workspace" ));
+    } else {
+      ctx->blockstore = fd_blockstore_join( &ctx->blockstore_ljoin, fd_topo_obj_laddr( topo, blockstore_obj_id ) );
+      FD_TEST( ctx->blockstore!=NULL );
+    }
   }
 
-  ctx->blockstore = fd_blockstore_join( &ctx->blockstore_ljoin, fd_topo_obj_laddr( topo, blockstore_obj_id ) );
-  FD_TEST( ctx->blockstore!=NULL );
-
   FD_LOG_NOTICE(( "repair starting" ));
 
   /* Repair set up */

src/discof/repair/fd_repair_tile.c

@@ -14,18 +14,16 @@ static inline ulong
 generate_stake_weight_msg( fd_exec_slot_ctx_t * slot_ctx,
                            fd_spad_t          * runtime_spad,
                            ulong                epoch,
+                           fd_vote_accounts_global_t const * vote_accounts,
                            ulong              * stake_weight_msg_out ) {
   /* This function needs to be completely rewritten for SIMD-0180.
      For now it's a hack that sends old data (pre SIMD-0180) in the new format. */
 
   fd_stake_weight_msg_t *           stake_weight_msg = (fd_stake_weight_msg_t *)fd_type_pun( stake_weight_msg_out );
   fd_vote_stake_weight_t *          stake_weights    = stake_weight_msg->weights;
-  fd_vote_accounts_global_t const * vote_accounts    = fd_bank_epoch_stakes_locking_query( slot_ctx->bank );
   ulong                             staked_cnt       = fd_stake_weights_by_node( vote_accounts,
                                                                            stake_weights,
                                                                            runtime_spad );
-  fd_bank_epoch_stakes_end_locking_query( slot_ctx->bank );
-
   fd_epoch_schedule_t const * epoch_schedule = fd_bank_epoch_schedule_query( slot_ctx->bank );
 
   stake_weight_msg->epoch          = epoch;

src/discof/replay/fd_exec.h

@@ -300,7 +300,7 @@ publish_stake_weights( fd_replay_tile_ctx_t * ctx,
   if( epoch_stakes_root!=NULL ) {
     ulong * stake_weights_msg = fd_chunk_to_laddr( ctx->stake_out->mem, ctx->stake_out->chunk );
     ulong epoch = fd_slot_to_leader_schedule_epoch( epoch_schedule, fd_bank_slot_get( slot_ctx->bank ) );
-    ulong stake_weights_sz = generate_stake_weight_msg( slot_ctx, ctx->runtime_spad, epoch - 1, stake_weights_msg );
+    ulong stake_weights_sz = generate_stake_weight_msg( slot_ctx, ctx->runtime_spad, epoch - 1, epoch_stakes, stake_weights_msg );
     ulong stake_weights_sig = 4UL;
     fd_stem_publish( stem, 0UL, stake_weights_sig, ctx->stake_out->chunk, stake_weights_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
     ctx->stake_out->chunk = fd_dcache_compact_next( ctx->stake_out->chunk, stake_weights_sz, ctx->stake_out->chunk0, ctx->stake_out->wmark );
@@ -315,7 +315,7 @@ publish_stake_weights( fd_replay_tile_ctx_t * ctx,
   if( next_epoch_stakes_root!=NULL ) {
     ulong * stake_weights_msg = fd_chunk_to_laddr( ctx->stake_out->mem, ctx->stake_out->chunk );
     ulong   epoch             = fd_slot_to_leader_schedule_epoch( epoch_schedule, fd_bank_slot_get( slot_ctx->bank ) ); /* epoch */
-    ulong stake_weights_sz = generate_stake_weight_msg( slot_ctx, ctx->runtime_spad, epoch, stake_weights_msg );
+    ulong stake_weights_sz = generate_stake_weight_msg( slot_ctx, ctx->runtime_spad, epoch, next_epoch_stakes, stake_weights_msg );
     ulong stake_weights_sig = 4UL;
     fd_stem_publish( stem, 0UL, stake_weights_sig, ctx->stake_out->chunk, stake_weights_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
     ctx->stake_out->chunk = fd_dcache_compact_next( ctx->stake_out->chunk, stake_weights_sz, ctx->stake_out->chunk0, ctx->stake_out->wmark );