Change getHealth to compare optimistically confirmed slots

core/src/validator.rs

@@ -980,7 +980,6 @@ impl Validator {
                 ledger_path,
                 config.validator_exit.clone(),
                 exit.clone(),
-                config.known_validators.clone(),
                 rpc_override_health_check.clone(),
                 startup_verification_complete,
                 optimistically_confirmed_bank.clone(),

docs/src/api/http.md

@@ -156,11 +156,9 @@ health-check mechanism for use by load balancers or other network
 infrastructure. This request will always return a HTTP 200 OK response with a body of
 "ok", "behind" or "unknown" based on the following conditions:
 
-1. If one or more `--known-validator` arguments are provided to `solana-validator` - "ok" is returned
-   when the node has within `HEALTH_CHECK_SLOT_DISTANCE` slots of the highest
-   known validator, otherwise "behind". "unknown" is returned when no slot
-   information from known validators is not yet available.
-2. "ok" is always returned if no known validators are provided.
+1. If the node is within `HEALTH_CHECK_SLOT_DISTANCE` slots from the latest cluster confirmed slot, "ok" is returned.
+2. If the node is behind more than `HEALTH_CHECK_SLOT_DISTANCE` slots from the latest cluster confirmed slot, an error is returned that will contain more information about far behind the node is.
+3. If the node is unable to determine its' health, an error is returned.
 
 ## JSON RPC API Reference
 

docs/src/api/methods/_getHealth.mdx

@@ -12,13 +12,8 @@ import {
 
 ## getHealth
 
-Returns the current health of the node.
-
-:::caution
-If one or more `--known-validator` arguments are provided to `solana-validator` - "ok" is returned
-when the node has within `HEALTH_CHECK_SLOT_DISTANCE` slots of the highest known validator,
-otherwise an error is returned. "ok" is always returned if no known validators are provided.
-:::
+Returns the current health of the node. A healthy node is one that is within
+`HEALTH_CHECK_SLOT_DISTANCE` slots of the latest cluster confirmed slot.
 
 <DocSideBySide>
 <CodeParams>

rpc/src/rpc.rs

@@ -374,31 +374,33 @@ impl JsonRpcRequestProcessor {
         );
 
         let leader_schedule_cache = Arc::new(LeaderScheduleCache::new_from_bank(&bank));
+        let startup_verification_complete = Arc::clone(bank.get_startup_verification_complete());
+        let slot = bank.slot();
+        let optimistically_confirmed_bank =
+            Arc::new(RwLock::new(OptimisticallyConfirmedBank { bank }));
         Self {
             config: JsonRpcConfig::default(),
             snapshot_config: None,
             bank_forks,
             block_commitment_cache: Arc::new(RwLock::new(BlockCommitmentCache::new(
                 HashMap::new(),
                 0,
-                CommitmentSlots::new_from_slot(bank.slot()),
+                CommitmentSlots::new_from_slot(slot),
             ))),
-            blockstore,
+            blockstore: Arc::clone(&blockstore),
             validator_exit: create_validator_exit(exit.clone()),
             health: Arc::new(RpcHealth::new(
-                cluster_info.clone(),
-                None,
+                Arc::clone(&optimistically_confirmed_bank),
+                blockstore,
                 0,
                 exit,
-                Arc::clone(bank.get_startup_verification_complete()),
+                startup_verification_complete,
             )),
             cluster_info,
             genesis_hash,
             transaction_sender: Arc::new(Mutex::new(sender)),
             bigtable_ledger_storage: None,
-            optimistically_confirmed_bank: Arc::new(RwLock::new(OptimisticallyConfirmedBank {
-                bank,
-            })),
+            optimistically_confirmed_bank,
             largest_accounts_cache: Arc::new(RwLock::new(LargestAccountsCache::new(30))),
             max_slots: Arc::new(MaxSlots::default()),
             leader_schedule_cache,
@@ -4787,6 +4789,8 @@ pub mod tests {
             // note that this means that slot 0 will always be considered complete
             let max_complete_transaction_status_slot = Arc::new(AtomicU64::new(0));
             let max_complete_rewards_slot = Arc::new(AtomicU64::new(0));
+            let optimistically_confirmed_bank =
+                OptimisticallyConfirmedBank::locked_from_bank_forks_root(&bank_forks);
 
             let meta = JsonRpcRequestProcessor::new(
                 config,
@@ -4795,11 +4799,11 @@ pub mod tests {
                 block_commitment_cache.clone(),
                 blockstore.clone(),
                 validator_exit,
-                RpcHealth::stub(),
+                RpcHealth::stub(optimistically_confirmed_bank.clone(), blockstore.clone()),
                 cluster_info,
                 Hash::default(),
                 None,
-                OptimisticallyConfirmedBank::locked_from_bank_forks_root(&bank_forks),
+                optimistically_confirmed_bank,
                 Arc::new(RwLock::new(LargestAccountsCache::new(30))),
                 max_slots.clone(),
                 Arc::new(LeaderScheduleCache::new_from_bank(&bank)),
@@ -6398,7 +6402,11 @@ pub mod tests {
         let blockstore = Arc::new(Blockstore::open(&ledger_path).unwrap());
         let block_commitment_cache = Arc::new(RwLock::new(BlockCommitmentCache::default()));
         let (bank_forks, mint_keypair, ..) = new_bank_forks();
-        let health = RpcHealth::stub();
+        let optimistically_confirmed_bank =
+            OptimisticallyConfirmedBank::locked_from_bank_forks_root(&bank_forks);
+        let health = RpcHealth::stub(optimistically_confirmed_bank.clone(), blockstore.clone());
+        // Mark the node as healthy to start
+        health.stub_set_health_status(Some(RpcHealthStatus::Ok));
 
         // Freeze bank 0 to prevent a panic in `run_transaction_simulation()`
         bank_forks.write().unwrap().get(0).unwrap().freeze();
@@ -6429,7 +6437,7 @@ pub mod tests {
             cluster_info,
             Hash::default(),
             None,
-            OptimisticallyConfirmedBank::locked_from_bank_forks_root(&bank_forks),
+            optimistically_confirmed_bank,
             Arc::new(RwLock::new(LargestAccountsCache::new(30))),
             Arc::new(MaxSlots::default()),
             Arc::new(LeaderScheduleCache::default()),
@@ -6690,18 +6698,20 @@ pub mod tests {
             .my_contact_info()
             .tpu(connection_cache.protocol())
             .unwrap();
+        let optimistically_confirmed_bank =
+            OptimisticallyConfirmedBank::locked_from_bank_forks_root(&bank_forks);
         let (request_processor, receiver) = JsonRpcRequestProcessor::new(
             JsonRpcConfig::default(),
             None,
             bank_forks.clone(),
             block_commitment_cache,
-            blockstore,
+            blockstore.clone(),
             validator_exit,
-            RpcHealth::stub(),
+            RpcHealth::stub(optimistically_confirmed_bank.clone(), blockstore),
             cluster_info,
             Hash::default(),
             None,
-            OptimisticallyConfirmedBank::locked_from_bank_forks_root(&bank_forks),
+            optimistically_confirmed_bank,
             Arc::new(RwLock::new(LargestAccountsCache::new(30))),
             Arc::new(MaxSlots::default()),
             Arc::new(LeaderScheduleCache::default()),
@@ -8327,9 +8337,9 @@ pub mod tests {
             None,
             bank_forks.clone(),
             block_commitment_cache,
-            blockstore,
+            blockstore.clone(),
             validator_exit,
-            RpcHealth::stub(),
+            RpcHealth::stub(optimistically_confirmed_bank.clone(), blockstore.clone()),
             cluster_info,
             Hash::default(),
             None,