replay, sched: verify PoH

Author: yufeng-jump

src/app/firedancer/config/default.toml

@@ -766,7 +766,7 @@ user = ""
     # much parallelism is possible.
     #
     # More exec tiles can allow the validator to replay through blocks
-    # faster, subject to Amdahl's law.  Empirically, we've found 8 exec
+    # faster, subject to Amdahl's law.  Empirically, we've found 10 exec
     # tiles to achieve near optimal replay speed under current
     # `mainnet-beta` conditions.
     #
@@ -775,7 +775,7 @@ user = ""
     # to be highly concurrent, most of the time account writeback can be
     # done in parallel without blocking.  Multiple exec tiles exploit this
     # parallelism supported by the accounts database.
-    exec_tile_count = 8
+    exec_tile_count = 10
 
     # How many shred tiles to run.  Should be set to 1.  This is
     # configurable and designed to scale out for future network

src/discof/exec/fd_exec_tile.c

@@ -2,6 +2,7 @@
 #include "generated/fd_exec_tile_seccomp.h"
 
 #include "../../util/pod/fd_pod_format.h"
+#include "../../ballet/sha256/fd_sha256.h" /* fd_sha256_hash_32_repeated */
 #include "../../discof/replay/fd_exec.h"
 #include "../../flamenco/capture/fd_capture_ctx.h"
 #include "../../flamenco/runtime/fd_bank.h"
@@ -145,7 +146,7 @@ returnable_frag( fd_exec_tile_ctx_t * ctx,
         fd_exec_txn_exec_msg_t * msg = fd_chunk_to_laddr( ctx->replay_in->mem, chunk );
         ctx->bank = fd_banks_bank_query( ctx->banks, msg->bank_idx );
         FD_TEST( ctx->bank );
-        ctx->txn_in.txn           = &msg->txn;
+        ctx->txn_in.txn           = msg->txn;
         ctx->txn_in.exec_accounts = &ctx->exec_accounts;
 
         /* Set the capture txn index from the message so account updates
@@ -181,7 +182,7 @@ returnable_frag( fd_exec_tile_ctx_t * ctx,
       }
       case FD_EXEC_TT_TXN_SIGVERIFY: {
         fd_exec_txn_sigverify_msg_t * msg = fd_chunk_to_laddr( ctx->replay_in->mem, chunk );
-        int res = fd_executor_txn_verify( &msg->txn, ctx->sha_lj );
+        int res = fd_executor_txn_verify( msg->txn, ctx->sha_lj );
         fd_exec_task_done_msg_t * out_msg = fd_chunk_to_laddr( ctx->exec_replay_out->mem, ctx->exec_replay_out->chunk );
         out_msg->bank_idx               = msg->bank_idx;
         out_msg->txn_sigverify->txn_idx = msg->txn_idx;
@@ -190,6 +191,17 @@ returnable_frag( fd_exec_tile_ctx_t * ctx,
         ctx->exec_replay_out->chunk = fd_dcache_compact_next( ctx->exec_replay_out->chunk, sizeof(*out_msg), ctx->exec_replay_out->chunk0, ctx->exec_replay_out->wmark );
         break;
       }
+      case FD_EXEC_TT_POH_HASH: {
+        fd_exec_poh_hash_msg_t * msg = fd_chunk_to_laddr( ctx->replay_in->mem, chunk );
+        fd_exec_task_done_msg_t * out_msg = fd_chunk_to_laddr( ctx->exec_replay_out->mem, ctx->exec_replay_out->chunk );
+        out_msg->bank_idx           = msg->bank_idx;
+        out_msg->poh_hash->mblk_idx = msg->mblk_idx;
+        out_msg->poh_hash->hashcnt  = msg->hashcnt;
+        fd_sha256_hash_32_repeated( msg->hash, out_msg->poh_hash->hash, msg->hashcnt );
+        fd_stem_publish( stem, ctx->exec_replay_out->idx, (FD_EXEC_TT_POH_HASH<<32)|ctx->tile_idx, ctx->exec_replay_out->chunk, sizeof(*out_msg), 0UL, 0UL, 0UL );
+        ctx->exec_replay_out->chunk = fd_dcache_compact_next( ctx->exec_replay_out->chunk, sizeof(*out_msg), ctx->exec_replay_out->chunk0, ctx->exec_replay_out->wmark );
+        break;
+      }
       default: FD_LOG_CRIT(( "unexpected signature %lu", sig ));
     }
   } else FD_LOG_CRIT(( "invalid in_idx %lu", in_idx ));

src/discof/replay/fd_exec.h

@@ -10,7 +10,7 @@
 #define FD_EXEC_TT_TXN_EXEC      (1UL) /* Transaction execution. */
 #define FD_EXEC_TT_TXN_SIGVERIFY (2UL) /* Transaction sigverify. */
 #define FD_EXEC_TT_LTHASH        (3UL) /* Account lthash. */
-#define FD_EXEC_TT_POH_VERIFY    (4UL) /* PoH hash verification. */
+#define FD_EXEC_TT_POH_HASH      (4UL) /* PoH hashing. */
 
 /* Sent from the replay tile to the exec tiles.  These describe one of
    several types of tasks for an exec tile.  An idx to the bank in the
@@ -20,7 +20,7 @@
 struct fd_exec_txn_exec_msg {
   ulong      bank_idx;
   ulong      txn_idx;
-  fd_txn_p_t txn;
+  fd_txn_p_t txn[ 1 ];
 
   /* Used currently by solcap to maintain ordering of messages
      this will change to using txn sigs eventually */
@@ -31,13 +31,22 @@ typedef struct fd_exec_txn_exec_msg fd_exec_txn_exec_msg_t;
 struct fd_exec_txn_sigverify_msg {
   ulong      bank_idx;
   ulong      txn_idx;
-  fd_txn_p_t txn;
+  fd_txn_p_t txn[ 1 ];
 };
 typedef struct fd_exec_txn_sigverify_msg fd_exec_txn_sigverify_msg_t;
 
+struct fd_exec_poh_hash_msg {
+  ulong     bank_idx;
+  ulong     mblk_idx;
+  ulong     hashcnt;
+  fd_hash_t hash[ 1 ];
+};
+typedef struct fd_exec_poh_hash_msg fd_exec_poh_hash_msg_t;
+
 union fd_exec_task_msg {
   fd_exec_txn_exec_msg_t      txn_exec;
   fd_exec_txn_sigverify_msg_t txn_sigverify;
+  fd_exec_poh_hash_msg_t      poh_hash;
 };
 typedef union fd_exec_task_msg fd_exec_task_msg_t;
 
@@ -63,11 +72,19 @@ struct fd_exec_txn_sigverify_done_msg {
 };
 typedef struct fd_exec_txn_sigverify_done_msg fd_exec_txn_sigverify_done_msg_t;
 
+struct fd_exec_poh_hash_done_msg {
+  ulong     mblk_idx;
+  ulong     hashcnt;
+  fd_hash_t hash[ 1 ];
+};
+typedef struct fd_exec_poh_hash_done_msg fd_exec_poh_hash_done_msg_t;
+
 struct fd_exec_task_done_msg {
   ulong bank_idx;
   union {
     fd_exec_txn_exec_done_msg_t      txn_exec[ 1 ];
     fd_exec_txn_sigverify_done_msg_t txn_sigverify[ 1 ];
+    fd_exec_poh_hash_done_msg_t      poh_hash[ 1 ];
   };
 };
 typedef struct fd_exec_task_done_msg fd_exec_task_done_msg_t;

src/discof/replay/fd_replay_tile.c

@@ -606,22 +606,21 @@ replay_block_start( fd_replay_tile_t *  ctx,
 
   fd_bank_has_identity_vote_set( bank, 0 );
 
-  /* Set the tick height. */
-  fd_bank_tick_height_set( bank, fd_bank_max_tick_height_get( bank ) );
-
   /* Update block height. */
   fd_bank_block_height_set( bank, fd_bank_block_height_get( bank ) + 1UL );
 
-  ulong * max_tick_height = fd_bank_max_tick_height_modify( bank );
-  ulong   ticks_per_slot  = fd_bank_ticks_per_slot_get( bank );
-  if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS != fd_runtime_compute_max_tick_height( ticks_per_slot, slot, max_tick_height ) ) ) {
-    FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, ticks_per_slot ));
-  }
-
   int is_epoch_boundary = 0;
   fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
   if( FD_UNLIKELY( is_epoch_boundary ) ) publish_stake_weights( ctx, stem, bank, 1 );
 
+  ulong max_tick_height;
+  if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( fd_bank_ticks_per_slot_get( parent_bank ), slot, &max_tick_height ) ) ) {
+    FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, fd_bank_ticks_per_slot_get( bank ) ));
+  }
+  fd_bank_max_tick_height_set( bank, max_tick_height );
+  fd_bank_tick_height_set( bank, fd_bank_max_tick_height_get( parent_bank ) ); /* The parent's max tick height is our starting tick height. */
+  fd_sched_set_poh_params( ctx->sched, bank->idx, fd_bank_tick_height_get( bank ), fd_bank_max_tick_height_get( bank ), fd_bank_hashes_per_tick_get( bank ), fd_bank_poh_query( parent_bank ) );
+
   return bank;
 }
 
@@ -1524,7 +1523,7 @@ dispatch_task( fd_replay_tile_t *  ctx,
 
       fd_replay_out_link_t *   exec_out = ctx->exec_out;
       fd_exec_txn_exec_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
-      memcpy( &exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
+      memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
       exec_msg->bank_idx = task->txn_exec->bank_idx;
       exec_msg->txn_idx  = task->txn_exec->txn_idx;
       if( FD_UNLIKELY( ctx->capture_ctx ) ) {
@@ -1542,13 +1541,27 @@ dispatch_task( fd_replay_tile_t *  ctx,
 
       fd_replay_out_link_t *        exec_out = ctx->exec_out;
       fd_exec_txn_sigverify_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
-      memcpy( &exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
+      memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
       exec_msg->bank_idx = task->txn_sigverify->bank_idx;
       exec_msg->txn_idx  = task->txn_sigverify->txn_idx;
       fd_stem_publish( stem, exec_out->idx, (FD_EXEC_TT_TXN_SIGVERIFY<<32) | task->txn_sigverify->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
       exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
       break;
     };
+    case FD_SCHED_TT_POH_HASH: {
+      fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->poh_hash->bank_idx );
+      bank->refcnt++;
+
+      fd_replay_out_link_t *   exec_out = ctx->exec_out;
+      fd_exec_poh_hash_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
+      exec_msg->bank_idx = task->poh_hash->bank_idx;
+      exec_msg->mblk_idx = task->poh_hash->mblk_idx;
+      exec_msg->hashcnt  = task->poh_hash->hashcnt;
+      memcpy( exec_msg->hash, task->poh_hash->hash, sizeof(fd_hash_t) );
+      fd_stem_publish( stem, exec_out->idx, (FD_EXEC_TT_POH_HASH<<32) | task->poh_hash->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
+      exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
+      break;
+    };
     default: {
       FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
     }
@@ -1573,22 +1586,29 @@ replay( fd_replay_tile_t *  ctx,
   switch( task->task_type ) {
     case FD_SCHED_TT_BLOCK_START: {
       replay_block_start( ctx, stem, task->block_start->bank_idx, task->block_start->parent_bank_idx, task->block_start->slot );
-      fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX );
+      fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX, NULL );
       break;
     }
     case FD_SCHED_TT_BLOCK_END: {
       fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->block_end->bank_idx );
       if( FD_LIKELY( !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) replay_block_finalize( ctx, stem, bank );
-      fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX );
+      fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX, NULL );
       break;
     }
     case FD_SCHED_TT_TXN_EXEC:
-    case FD_SCHED_TT_TXN_SIGVERIFY: {
+    case FD_SCHED_TT_TXN_SIGVERIFY:
+    case FD_SCHED_TT_POH_HASH: {
       /* Likely/common case: we have a transaction we actually need to
          execute. */
       dispatch_task( ctx, stem, task );
       break;
     }
+    case FD_SCHED_TT_MARK_DEAD: {
+      fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->mark_dead->bank_idx );
+      publish_slot_dead( ctx, stem, bank );
+      fd_banks_mark_bank_dead( ctx->banks, bank );
+      break;
+    }
     default: {
       FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
     }
@@ -1957,7 +1977,8 @@ process_exec_task_done( fd_replay_tile_t *        ctx,
       if( FD_UNLIKELY( (bank->flags&FD_BANK_FLAGS_DEAD) && bank->refcnt==0UL ) ) {
         fd_banks_mark_bank_frozen( ctx->banks, bank );
       }
-      fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, msg->txn_exec->txn_idx, exec_tile_idx );
+      int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, msg->txn_exec->txn_idx, exec_tile_idx, NULL );
+      FD_TEST( res==0 );
       break;
     }
     case FD_EXEC_TT_TXN_SIGVERIFY: {
@@ -1972,7 +1993,17 @@ process_exec_task_done( fd_replay_tile_t *        ctx,
       if( FD_UNLIKELY( (bank->flags&FD_BANK_FLAGS_DEAD) && bank->refcnt==0UL ) ) {
         fd_banks_mark_bank_frozen( ctx->banks, bank );
       }
-      fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, msg->txn_sigverify->txn_idx, exec_tile_idx );
+      int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, msg->txn_sigverify->txn_idx, exec_tile_idx, NULL );
+      FD_TEST( res==0 );
+      break;
+    }
+    case FD_EXEC_TT_POH_HASH: {
+      int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_POH_HASH, ULONG_MAX, exec_tile_idx, msg->poh_hash );
+      if( FD_UNLIKELY( res<0 && !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) {
+        publish_slot_dead( ctx, stem, bank );
+        fd_banks_mark_bank_dead( ctx->banks, bank );
+      }
+      if( FD_UNLIKELY( (bank->flags&FD_BANK_FLAGS_DEAD) && bank->refcnt==0UL ) ) fd_banks_mark_bank_frozen( ctx->banks, bank );
       break;
     }
     default: FD_LOG_CRIT(( "unexpected sig 0x%lx", sig ));