Skip to content

repair: worker pooled async signing and ping-pong optimization #5924

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 11, 2025
Merged

repair: worker pooled async signing and ping-pong optimization #5924

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 20 additions & 9 deletions src/app/firedancer-dev/commands/repair.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,7 @@ repair_topo( config_t * config ) {
ulong net_tile_cnt = config->layout.net_tile_count;
ulong shred_tile_cnt = config->layout.shred_tile_count;
ulong quic_tile_cnt = config->layout.quic_tile_count;
ulong sign_tile_cnt = config->firedancer.layout.sign_tile_count;

fd_topo_t * topo = { fd_topob_new( &config->topo, config->name ) };
topo->max_page_size = fd_cstr_to_shmem_page_sz( config->hugetlbfs.max_page_size );
Expand Down Expand Up @@ -132,12 +133,17 @@ repair_topo( config_t * config ) {
/**/ fd_topob_link( topo, "gossip_net", "net_gossip", config->net.ingress_buffer_size, FD_NET_MTU, 1UL );

/**/ fd_topob_link( topo, "repair_net", "net_repair", config->net.ingress_buffer_size, FD_NET_MTU, 1UL );
/**/ fd_topob_link( topo, "repair_sign", "repair_sign", 128UL, 2048UL, 1UL );
FOR(shred_tile_cnt) fd_topob_link( topo, "shred_repair", "shred_repair", pending_fec_shreds_depth, FD_SHRED_REPAIR_MTU, 2UL );

FOR(shred_tile_cnt) fd_topob_link( topo, "shred_repair", "shred_repair", pending_fec_shreds_depth, FD_SHRED_REPAIR_MTU, 2UL /* at most 2 msgs per after_frag */ );

FOR(shred_tile_cnt) fd_topob_link( topo, "repair_shred", "shred_repair", pending_fec_shreds_depth, sizeof(fd_ed25519_sig_t), 1UL );
/**/ fd_topob_link( topo, "sign_repair", "sign_repair", 128UL, 64UL, 1UL );
/**/ fd_topob_link( topo, "repair_repla", "repair_repla", 65536UL, sizeof(fd_reasm_fec_t), 1UL );

/**/ fd_topob_link( topo, "ping_sign", "repair_sign", 128UL, 2048UL, 1UL );
/**/ fd_topob_link( topo, "sign_ping", "sign_repair", 128UL, sizeof(fd_ed25519_sig_t), 1UL );
FOR(sign_tile_cnt-1) fd_topob_link( topo, "repair_sign", "repair_sign", 128UL, 2048UL, 1UL );
FOR(sign_tile_cnt-1) fd_topob_link( topo, "sign_repair", "sign_repair", 1024UL, sizeof(fd_ed25519_sig_t), 1UL );

/**/ fd_topob_link( topo, "repair_repla", "repair_repla", 65536UL, sizeof(fd_reasm_fec_t), 1UL );
/**/ fd_topob_link( topo, "poh_shred", "poh_shred", 16384UL, USHORT_MAX, 1UL );

/**/ fd_topob_link( topo, "send_txns", "send_txns", 128UL, FD_TXN_MTU, 1UL );
Expand Down Expand Up @@ -181,7 +187,7 @@ repair_topo( config_t * config ) {

/* topo, tile_name, tile_wksp, metrics_wksp, cpu_idx, is_agave, uses_keyswitch */
FOR(shred_tile_cnt) fd_topob_tile( topo, "shred", "shred", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1 );
/**/ fd_topob_tile( topo, "sign", "sign", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1 );
FOR(sign_tile_cnt) fd_topob_tile( topo, "sign", "sign", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1 );
/**/ fd_topob_tile( topo, "metric", "metric", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0 );
/**/ fd_topob_tile( topo, "gossip", "gossip", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0 );
fd_topo_tile_t * repair_tile = fd_topob_tile( topo, "repair", "repair", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0 );
Expand Down Expand Up @@ -303,12 +309,17 @@ repair_topo( config_t * config ) {
fd_topob_tile_in( topo, "repair", 0UL, "metric_in", "snap_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
FOR(shred_tile_cnt) fd_topob_tile_in( topo, "repair", 0UL, "metric_in", "shred_repair", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );

/**/ fd_topob_tile_in( topo, "sign", 0UL, "metric_in", "repair_sign", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );
/**/ fd_topob_tile_out( topo, "repair", 0UL, "repair_sign", 0UL );
/**/ fd_topob_tile_in( topo, "repair", 0UL, "metric_in", "sign_repair", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_UNPOLLED );
/**/ fd_topob_tile_in( topo, "sign", 0UL, "metric_in", "ping_sign", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
/**/ fd_topob_tile_out( topo, "repair", 0UL, "ping_sign", 0UL );
/**/ fd_topob_tile_out( topo, "repair", 0UL, "repair_repla", 0UL );
FOR(shred_tile_cnt) fd_topob_tile_out( topo, "repair", 0UL, "repair_shred", i );
/**/ fd_topob_tile_out( topo, "sign", 0UL, "sign_repair", 0UL );
/**/ fd_topob_tile_out( topo, "sign", 0UL, "sign_ping", 0UL );

FOR(sign_tile_cnt-1) fd_topob_tile_out( topo, "repair", 0UL, "repair_sign", i );
FOR(sign_tile_cnt-1) fd_topob_tile_in ( topo, "sign", i+1, "metric_in", "repair_sign", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
FOR(sign_tile_cnt-1) fd_topob_tile_out( topo, "sign", i+1, "sign_repair", i );
FOR(sign_tile_cnt-1) fd_topob_tile_in ( topo, "repair", 0UL, "metric_in", "sign_repair", i, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );
/**/ fd_topob_tile_in ( topo, "repair", 0UL, "metric_in", "sign_ping", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_UNPOLLED );

if( 1 ) {
fd_topob_wksp( topo, "scap" );
Expand Down
7 changes: 7 additions & 0 deletions src/app/firedancer/config/default.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -691,6 +691,13 @@ user = ""
# very high TPS rates because the cluster size will be very small.
shred_tile_count = 1

# How many sign tiles to run. Should be set >= 2. This is
# configurable and horizontally scales repair request signing.
# One tile is reserved for synchronous signing across all tiles.
# The remaining tiles distribute the workload of signing repair
# requests.
sign_tile_count = 2

# All memory that will be used in Firedancer is pre-allocated in two
# kinds of pages: huge and gigantic. Huge pages are 2 MiB and gigantic
# pages are 1 GiB. This is done to prevent TLB misses which can have a
Expand Down
Loading
Loading