tower: add hard fork detector

mmcgee-jump · lidatong · commit 8c70efef2310 · 2025-11-19T10:50:11.000-06:00
diff --git a/src/app/firedancer-dev/commands/backtest.c b/src/app/firedancer-dev/commands/backtest.c
@@ -313,13 +313,6 @@ backtest_topo( config_t * config ) {
   FOR(exec_tile_cnt) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "exec", i ) ], banks_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
   FD_TEST( fd_pod_insertf_ulong( topo->props, banks_obj->id, "banks" ) );
 
-  /* bank_hash_cmp_obj shared by replay and exec tiles */
-  fd_topob_wksp( topo, "bh_cmp" );
-  fd_topo_obj_t * bank_hash_cmp_obj = setup_topo_bank_hash_cmp( topo, "bh_cmp" );
-  fd_topob_tile_uses( topo, replay_tile, bank_hash_cmp_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
-  FOR(exec_tile_cnt) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "exec", i ) ], bank_hash_cmp_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
-  FD_TEST( fd_pod_insertf_ulong( topo->props, bank_hash_cmp_obj->id, "bh_cmp" ) );
-
   /* txncache_obj, busy_obj and poh_slot_obj only by replay tile */
   fd_topob_wksp( topo, "txncache"    );
   fd_topob_wksp( topo, "bank_busy"   );
@@ -358,7 +351,6 @@ backtest_topo( config_t * config ) {
     fd_topo_configure_tile( tile, config );
 
     if( !strcmp( tile->name, "replay" ) ) {
-      tile->replay.enable_bank_hash_cmp = 0;
       tile->replay.enable_features_cnt = config->tiles.replay.enable_features_cnt;
       for( ulong i = 0; i < tile->replay.enable_features_cnt; i++ ) {
         strncpy( tile->replay.enable_features[i], config->tiles.replay.enable_features[i], sizeof(tile->replay.enable_features[i]) );
diff --git a/src/app/firedancer-dev/main.c b/src/app/firedancer-dev/main.c
@@ -24,7 +24,6 @@ extern fd_topo_obj_callbacks_t fd_obj_cb_fec_sets;
 extern fd_topo_obj_callbacks_t fd_obj_cb_txncache;
 extern fd_topo_obj_callbacks_t fd_obj_cb_banks;
 extern fd_topo_obj_callbacks_t fd_obj_cb_funk;
-extern fd_topo_obj_callbacks_t fd_obj_cb_bank_hash_cmp;
 
 extern fd_topo_obj_callbacks_t fd_obj_cb_vinyl_meta;
 extern fd_topo_obj_callbacks_t fd_obj_cb_vinyl_meta_ele;
@@ -47,7 +46,6 @@ fd_topo_obj_callbacks_t * CALLBACKS[] = {
   &fd_obj_cb_txncache,
   &fd_obj_cb_banks,
   &fd_obj_cb_funk,
-  &fd_obj_cb_bank_hash_cmp,
   &fd_obj_cb_vinyl_meta,
   &fd_obj_cb_vinyl_meta_ele,
   &fd_obj_cb_vinyl_data,
diff --git a/src/app/firedancer/callbacks.c b/src/app/firedancer/callbacks.c
@@ -38,34 +38,6 @@ fd_topo_obj_callbacks_t fd_obj_cb_banks = {
   .new       = banks_new,
 };
 
-static ulong
-bh_cmp_footprint( fd_topo_t const *     topo,
-                  fd_topo_obj_t const * obj ) {
-  (void)topo; (void)obj;
-  return fd_bank_hash_cmp_footprint();
-}
-
-static ulong
-bh_cmp_align( fd_topo_t const *     topo,
-              fd_topo_obj_t const * obj ) {
-  (void)topo; (void)obj;
-  return fd_bank_hash_cmp_align();
-}
-
-static void
-bh_cmp_new( fd_topo_t const *     topo,
-                   fd_topo_obj_t const * obj ) {
-  (void)topo; (void)obj;
-  FD_TEST( fd_bank_hash_cmp_new( fd_topo_obj_laddr( topo, obj->id ) ) );
-}
-
-fd_topo_obj_callbacks_t fd_obj_cb_bank_hash_cmp = {
-  .name      = "bh_cmp",
-  .footprint = bh_cmp_footprint,
-  .align     = bh_cmp_align,
-  .new       = bh_cmp_new,
-};
-
 static ulong
 funk_align( fd_topo_t const *     topo,
             fd_topo_obj_t const * obj ) {
diff --git a/src/app/firedancer/config/default.toml b/src/app/firedancer/config/default.toml
@@ -1545,6 +1545,19 @@ user = ""
     # enabled during routine running of the validator.
     core_dump = true
 
+    # A hard fork occurs when our validator has some implementation
+    # difference in consensus rules and gets a different blockhash for
+    # a given slot than the rest of the cluster.
+    #
+    # Typically, we should assume that our validator is correct and keep
+    # running, but during development it can be useful to detect if we
+    # are on the minority hard fork and immediately exit or abort.
+    #
+    # If this option is true, the validator will immediately exit with
+    # an error if it detects a hard fork.  If false, the validator will
+    # log the hard fork event and continue running.
+    hard_fork_fatal = true
+
     # It can be convenient during development to use a network namespace
     # for running Firedancer.  This allows us to send packets at a local
     # Firedancer instance and have them go through more of the kernel
diff --git a/src/app/firedancer/main.c b/src/app/firedancer/main.c
@@ -21,7 +21,6 @@ extern fd_topo_obj_callbacks_t fd_obj_cb_fec_sets;
 extern fd_topo_obj_callbacks_t fd_obj_cb_txncache;
 extern fd_topo_obj_callbacks_t fd_obj_cb_banks;
 extern fd_topo_obj_callbacks_t fd_obj_cb_funk;
-extern fd_topo_obj_callbacks_t fd_obj_cb_bank_hash_cmp;
 
 fd_topo_obj_callbacks_t * CALLBACKS[] = {
   &fd_obj_cb_mcache,
@@ -39,7 +38,6 @@ fd_topo_obj_callbacks_t * CALLBACKS[] = {
   &fd_obj_cb_txncache,
   &fd_obj_cb_banks,
   &fd_obj_cb_funk,
-  &fd_obj_cb_bank_hash_cmp,
   NULL,
 };
 
diff --git a/src/app/firedancer/topology.c b/src/app/firedancer/topology.c
@@ -52,12 +52,6 @@ parse_ip_port( const char * name, const char * ip_port, fd_topo_ip_port_t *parse
     FD_LOG_ERR(( "could not parse port %s in [%s]", ip_end+1, name ));
 }
 
-fd_topo_obj_t *
-setup_topo_bank_hash_cmp( fd_topo_t * topo, char const * wksp_name ) {
-  fd_topo_obj_t * obj = fd_topob_obj( topo, "bh_cmp", wksp_name );
-  return obj;
-}
-
 fd_topo_obj_t *
 setup_topo_banks( fd_topo_t *  topo,
                   char const * wksp_name,
@@ -940,12 +934,6 @@ fd_topo_initialize( config_t * config ) {
     }
   }
 
-  /* TODO: This should not exist in production */
-  fd_topo_obj_t * bank_hash_cmp_obj = setup_topo_bank_hash_cmp( topo, "bh_cmp" );
-  /**/               fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "replay", 0UL ) ], bank_hash_cmp_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
-  FOR(exec_tile_cnt) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "exec",   i   ) ], bank_hash_cmp_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
-  FD_TEST( fd_pod_insertf_ulong( topo->props, bank_hash_cmp_obj->id, "bh_cmp" ) );
-
   ulong fec_set_cnt = shred_depth + config->tiles.shred.max_pending_shred_sets + 4UL;
   ulong fec_sets_sz = fec_set_cnt*sizeof(fd_shred34_t)*4; /* mirrors # of dcache entires in frankendancer */
   fd_topo_obj_t * fec_sets_obj = setup_topo_fec_sets( topo, "fec_sets", shred_tile_cnt*fec_sets_sz );
@@ -1185,7 +1173,6 @@ fd_topo_configure_tile( fd_topo_tile_t * tile,
     strncpy( tile->replay.identity_key_path, config->paths.identity_key, sizeof(tile->replay.identity_key_path) );
     tile->replay.ip_addr = config->net.ip_addr;
     strncpy( tile->replay.vote_account_path, config->paths.vote_account, sizeof(tile->replay.vote_account_path) );
-    tile->replay.enable_bank_hash_cmp = 1;
 
     tile->replay.capture_start_slot = config->capture.capture_start_slot;
     strncpy( tile->replay.solcap_capture, config->capture.solcap_capture, sizeof(tile->replay.solcap_capture) );
@@ -1212,6 +1199,7 @@ fd_topo_configure_tile( fd_topo_tile_t * tile,
 
   } else if( FD_UNLIKELY( !strcmp( tile->name, "tower" ) ) ) {
 
+    tile->tower.fork_fatal         = config->firedancer.development.hard_fork_fatal;
     tile->tower.max_live_slots     = config->firedancer.runtime.max_live_slots;
     tile->tower.max_lookahead_conf = config->tiles.tower.max_lookahead_conf;
     strncpy( tile->tower.identity_key, config->paths.identity_key, sizeof(tile->tower.identity_key) );
diff --git a/src/app/firedancer/topology.h b/src/app/firedancer/topology.h
@@ -13,9 +13,6 @@ FD_PROTOTYPES_BEGIN
 void
 fd_topo_initialize( fd_config_t * config );
 
-fd_topo_obj_t *
-setup_topo_bank_hash_cmp( fd_topo_t * topo, char const * wksp_name );
-
 fd_topo_obj_t *
 setup_topo_banks( fd_topo_t *  topo,
                   char const * wksp_name,
diff --git a/src/app/shared/fd_config.h b/src/app/shared/fd_config.h
@@ -161,6 +161,10 @@ struct fd_configf {
     uint full_effective_age_cancel_threshold;
   } snapshots;
 
+  struct {
+    int hard_fork_fatal;
+  } development;
+
   struct {
     ulong max_completed_shred_sets;
   } store;
diff --git a/src/app/shared/fd_config_parse.c b/src/app/shared/fd_config_parse.c
@@ -118,6 +118,8 @@ fd_config_extract_podf( uchar *        pod,
   CFG_POP      ( uint,   snapshots.max_incremental_snapshots_to_keep         );
   CFG_POP      ( uint,   snapshots.full_effective_age_cancel_threshold       );
 
+  CFG_POP      ( bool,   development.hard_fork_fatal                         );
+
   return config;
 }
 
diff --git a/src/app/shared/fd_tile_unit_test.c b/src/app/shared/fd_tile_unit_test.c
@@ -21,7 +21,6 @@ extern fd_topo_obj_callbacks_t fd_obj_cb_fec_sets;
 extern fd_topo_obj_callbacks_t fd_obj_cb_txncache;
 extern fd_topo_obj_callbacks_t fd_obj_cb_banks;
 extern fd_topo_obj_callbacks_t fd_obj_cb_funk;
-extern fd_topo_obj_callbacks_t fd_obj_cb_bank_hash_cmp;
 
 fd_topo_obj_callbacks_t * CALLBACKS[] = {
   &fd_obj_cb_mcache,
@@ -39,7 +38,6 @@ fd_topo_obj_callbacks_t * CALLBACKS[] = {
   &fd_obj_cb_txncache,
   &fd_obj_cb_banks,
   &fd_obj_cb_funk,
-  &fd_obj_cb_bank_hash_cmp,
   NULL,
 };
 
diff --git a/src/disco/topo/fd_topo.h b/src/disco/topo/fd_topo.h
@@ -387,8 +387,6 @@ struct fd_topo_tile {
       ulong enable_features_cnt;
       char  enable_features[ 16 ][ FD_BASE58_ENCODED_32_SZ ];
 
-      ulong enable_bank_hash_cmp;
-
       int   larger_max_cost_per_block;
 
       ulong capture_start_slot;
@@ -494,6 +492,7 @@ struct fd_topo_tile {
     } archiver;
 
     struct {
+      int   fork_fatal;
       ulong max_live_slots;
       ulong max_lookahead_conf;
       char  identity_key[ PATH_MAX ];
diff --git a/src/discof/exec/fd_exec_tile.c b/src/discof/exec/fd_exec_tile.c
@@ -28,7 +28,6 @@ typedef struct link_ctx {
 } link_ctx_t;
 
 typedef struct fd_exec_tile_ctx {
-
   ulong                 tile_idx;
 
   /* link-related data structures. */
@@ -39,8 +38,6 @@ typedef struct fd_exec_tile_ctx {
   fd_sha512_t           sha_mem[ FD_TXN_ACTUAL_SIG_MAX ];
   fd_sha512_t *         sha_lj[ FD_TXN_ACTUAL_SIG_MAX ];
 
-  fd_bank_hash_cmp_t *  bank_hash_cmp;
-
   /* Data structures related to managing and executing the transaction.
      The fd_txn_p_t is refreshed with every transaction and is sent
      from the dispatch/replay tile.  The fd_exec_txn_ctx_t * is a valid
@@ -261,19 +258,6 @@ unprivileged_init( fd_topo_t *      topo,
     FD_LOG_ERR(( "Failed to join banks" ));
   }
 
-  /********************************************************************/
-  /* bank hash cmp                                                    */
-  /********************************************************************/
-
-  ulong bank_hash_cmp_obj_id = fd_pod_queryf_ulong( topo->props, ULONG_MAX, "bh_cmp" );
-  if( FD_UNLIKELY( bank_hash_cmp_obj_id==ULONG_MAX ) ) {
-    FD_LOG_ERR(( "Could not find topology object for bank hash cmp" ));
-  }
-  ctx->bank_hash_cmp = fd_bank_hash_cmp_join( fd_topo_obj_laddr( topo, bank_hash_cmp_obj_id ) );
-  if( FD_UNLIKELY( !ctx->bank_hash_cmp ) ) {
-    FD_LOG_ERR(( "Failed to join bank hash cmp" ));
-  }
-
   void * shfunk = fd_topo_obj_laddr( topo, tile->exec.funk_obj_id );
   if( FD_UNLIKELY( !fd_funk_join( ctx->funk, shfunk ) ) ) {
     FD_LOG_CRIT(( "fd_funk_join(accdb) failed" ));
diff --git a/src/discof/replay/fd_replay_tile.c b/src/discof/replay/fd_replay_tile.c
@@ -159,8 +159,6 @@ struct fd_replay_tile {
 
   /* Replay state machine. */
   fd_sched_t *         sched;
-  uint                 enable_bank_hash_cmp:1;
-  fd_bank_hash_cmp_t * bank_hash_cmp;
   ulong                exec_cnt;
   fd_replay_out_link_t exec_out[ 1 ]; /* Sending work down to exec tiles */
 
@@ -767,35 +765,6 @@ replay_block_finalize( fd_replay_tile_t *  ctx,
   fd_hash_t const * bank_hash = fd_bank_bank_hash_query( bank );
   FD_TEST( bank_hash );
 
-  fd_bank_hash_cmp_t * bank_hash_cmp = ctx->bank_hash_cmp;
-  fd_bank_hash_cmp_lock( bank_hash_cmp );
-  fd_bank_hash_cmp_insert( bank_hash_cmp, fd_bank_slot_get( bank ), bank_hash, 1, 0 );
-
-  /* Try to move the bank hash comparison watermark forward */
-  for( ulong cmp_slot = bank_hash_cmp->watermark + 1; cmp_slot < fd_bank_slot_get( bank ); cmp_slot++ ) {
-    if( FD_UNLIKELY( !ctx->enable_bank_hash_cmp ) ) {
-      bank_hash_cmp->watermark = cmp_slot;
-      break;
-    }
-    int rc = fd_bank_hash_cmp_check( bank_hash_cmp, cmp_slot );
-    switch ( rc ) {
-      case -1:
-        /* Mismatch */
-        FD_LOG_CRIT(( "Bank hash mismatch on slot: %lu. Halting.", cmp_slot ));
-        break;
-      case 0:
-        /* Not ready */
-        break;
-      case 1:
-        /* Match*/
-        bank_hash_cmp->watermark = cmp_slot;
-        break;
-      default:;
-    }
-  }
-
-  fd_bank_hash_cmp_unlock( bank_hash_cmp );
-
   /* Must be last so we can measure completion time correctly, even
      though we could technically do this before the hash cmp and vote
      tower stuff. */
@@ -926,35 +895,6 @@ fini_leader_bank( fd_replay_tile_t *  ctx,
   fd_hash_t const * bank_hash  = fd_bank_bank_hash_query( ctx->leader_bank );
   FD_TEST( bank_hash );
 
-  fd_bank_hash_cmp_t * bank_hash_cmp = ctx->bank_hash_cmp;
-  fd_bank_hash_cmp_lock( bank_hash_cmp );
-  fd_bank_hash_cmp_insert( bank_hash_cmp, curr_slot, bank_hash, 1, 0 );
-
-  /* Try to move the bank hash comparison watermark forward */
-  for( ulong cmp_slot = bank_hash_cmp->watermark + 1; cmp_slot < curr_slot; cmp_slot++ ) {
-    if( FD_UNLIKELY( !ctx->enable_bank_hash_cmp ) ) {
-      bank_hash_cmp->watermark = cmp_slot;
-      break;
-    }
-    int rc = fd_bank_hash_cmp_check( bank_hash_cmp, cmp_slot );
-    switch ( rc ) {
-      case -1:
-        /* Mismatch */
-        FD_LOG_WARNING(( "Bank hash mismatch on slot: %lu. Halting.", cmp_slot ));
-        break;
-      case 0:
-        /* Not ready */
-        break;
-      case 1:
-        /* Match*/
-        bank_hash_cmp->watermark = cmp_slot;
-        break;
-      default:;
-    }
-  }
-
-  fd_bank_hash_cmp_unlock( bank_hash_cmp );
-
   publish_slot_completed( ctx, stem, ctx->leader_bank, 0, 1 /* is_leader */ );
 
   /* The reference on the bank is finally no longer needed. */
@@ -1089,21 +1029,6 @@ init_after_snapshot( fd_replay_tile_t * ctx ) {
     snapshot_slot = 0UL;
   }
 
-  /* Initialize consensus structures post-snapshot */
-
-  fd_vote_states_t const * vote_states = fd_bank_vote_states_locking_query( bank );
-
-  fd_bank_hash_cmp_t * bank_hash_cmp = ctx->bank_hash_cmp;
-
-  fd_vote_states_iter_t iter_[1];
-  for( fd_vote_states_iter_t * iter = fd_vote_states_iter_init( iter_, vote_states ); !fd_vote_states_iter_done( iter ); fd_vote_states_iter_next( iter ) ) {
-    fd_vote_state_ele_t const * vote_state = fd_vote_states_iter_ele( iter );
-    bank_hash_cmp->total_stake += vote_state->stake;
-  }
-  bank_hash_cmp->watermark = snapshot_slot;
-
-  fd_bank_vote_states_end_locking_query( bank );
-
   if( FD_UNLIKELY( ctx->capture_ctx ) ) fd_solcap_writer_flush( ctx->capture_ctx->capture );
 }
 
@@ -2481,13 +2406,6 @@ unprivileged_init( fd_topo_t *      topo,
   ctx->sched = fd_sched_join( fd_sched_new( sched_mem, tile->replay.max_live_slots, ctx->exec_cnt ), tile->replay.max_live_slots );
   FD_TEST( ctx->sched );
 
-  ctx->enable_bank_hash_cmp = !!tile->replay.enable_bank_hash_cmp;
-
-  ulong bank_hash_cmp_obj_id = fd_pod_query_ulong( topo->props, "bh_cmp", ULONG_MAX );
-  FD_TEST( bank_hash_cmp_obj_id!=ULONG_MAX );
-  ctx->bank_hash_cmp = fd_bank_hash_cmp_join( fd_bank_hash_cmp_new( fd_topo_obj_laddr( topo, bank_hash_cmp_obj_id ) ) );
-  FD_TEST( ctx->bank_hash_cmp );
-
   ctx->vote_tracker = fd_vote_tracker_join( fd_vote_tracker_new( vote_tracker_mem, 0UL ) );
   FD_TEST( ctx->vote_tracker );
 
diff --git a/src/discof/tower/Local.mk b/src/discof/tower/Local.mk
@@ -1,3 +1,3 @@
 ifdef FD_HAS_ALLOCA
-$(call add-objs,fd_tower_tile,fd_discof)
+$(call add-objs,fd_tower_tile fd_hard_fork_detector,fd_discof)
 endif
diff --git a/src/discof/tower/fd_hard_fork_detector.c b/src/discof/tower/fd_hard_fork_detector.c
diff --git a/src/discof/tower/fd_hard_fork_detector.h b/src/discof/tower/fd_hard_fork_detector.h
diff --git a/src/flamenco/runtime/Local.mk b/src/flamenco/runtime/Local.mk
diff --git a/src/flamenco/runtime/context/fd_exec_txn_ctx.h b/src/flamenco/runtime/context/fd_exec_txn_ctx.h
diff --git a/src/flamenco/runtime/fd_bank_hash_cmp.c b/src/flamenco/runtime/fd_bank_hash_cmp.c
diff --git a/src/flamenco/runtime/fd_bank_hash_cmp.h b/src/flamenco/runtime/fd_bank_hash_cmp.h
diff --git a/src/flamenco/runtime/program/fd_vote_program.c b/src/flamenco/runtime/program/fd_vote_program.c

Original file line number	Diff line number	Diff line change
`@@ -118,6 +118,8 @@ fd_config_extract_podf( uchar * pod,`
`118`	`118`	`CFG_POP ( uint, snapshots.max_incremental_snapshots_to_keep );`
`119`	`119`	`CFG_POP ( uint, snapshots.full_effective_age_cancel_threshold );`
`120`	`120`
	`121`	`+ CFG_POP ( bool, development.hard_fork_fatal );`
	`122`	`+`
`121`	`123`	`return config;`
`122`	`124`	`}`
`123`	`125`