Merge pull request #5544 from stacks-network/fix/5502

jcnelson · web-flow · commit c900752b5ca2 · 2024-12-18T22:20:45.000Z
Fix/5502
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -12,6 +12,8 @@ and this project adheres to the versioning scheme outlined in the [README.md](RE
 
 ### Changed
 
+- Nodes will assume that all PoX anchor blocks exist by default, and stall initial block download indefinitely to await their arrival (#5502)
+
 ## [3.1.0.0.1]
 
 ### Added
diff --git a/stackslib/src/chainstate/coordinator/mod.rs b/stackslib/src/chainstate/coordinator/mod.rs
@@ -198,6 +198,9 @@ pub trait BlockEventDispatcher {
 }
 
 pub struct ChainsCoordinatorConfig {
+    /// true: assume all anchor blocks are present, and block chain sync until they arrive
+    /// false: process sortitions in reward cycles without anchor blocks
+    pub assume_present_anchor_blocks: bool,
     /// true: use affirmation maps before 2.1
     /// false: only use affirmation maps in 2.1 or later
     pub always_use_affirmation_maps: bool,
@@ -209,8 +212,17 @@ pub struct ChainsCoordinatorConfig {
 impl ChainsCoordinatorConfig {
     pub fn new() -> ChainsCoordinatorConfig {
         ChainsCoordinatorConfig {
-            always_use_affirmation_maps: false,
+            always_use_affirmation_maps: true,
             require_affirmed_anchor_blocks: true,
+            assume_present_anchor_blocks: true,
+        }
+    }
+
+    pub fn test_new() -> ChainsCoordinatorConfig {
+        ChainsCoordinatorConfig {
+            always_use_affirmation_maps: false,
+            require_affirmed_anchor_blocks: false,
+            assume_present_anchor_blocks: false,
         }
     }
 }
@@ -700,7 +712,7 @@ impl<'a, T: BlockEventDispatcher, U: RewardSetProvider, B: BurnchainHeaderReader
             notifier: (),
             atlas_config,
             atlas_db: Some(atlas_db),
-            config: ChainsCoordinatorConfig::new(),
+            config: ChainsCoordinatorConfig::test_new(),
             burnchain_indexer,
             refresh_stacker_db: Arc::new(AtomicBool::new(false)),
             in_nakamoto_epoch: false,
@@ -2336,6 +2348,20 @@ impl<
                     panic!("BUG: no epoch defined at height {}", header.block_height)
                 });
 
+        if self.config.assume_present_anchor_blocks {
+            // anchor blocks are always assumed to be present in the chain history,
+            // so report its absence if we don't have it.
+            if let PoxAnchorBlockStatus::SelectedAndUnknown(missing_anchor_block, _) =
+                &rc_info.anchor_status
+            {
+                info!(
+                    "Currently missing PoX anchor block {}, which is assumed to be present",
+                    &missing_anchor_block
+                );
+                return Ok(Some(missing_anchor_block.clone()));
+            }
+        }
+
         if cur_epoch.epoch_id >= StacksEpochId::Epoch21 || self.config.always_use_affirmation_maps {
             // potentially have an anchor block, but only process the next reward cycle (and
             // subsequent reward cycles) with it if the prepare-phase block-commits affirm its
diff --git a/stackslib/src/config/mod.rs b/stackslib/src/config/mod.rs
@@ -1656,6 +1656,7 @@ pub struct NodeConfig {
     pub use_test_genesis_chainstate: Option<bool>,
     pub always_use_affirmation_maps: bool,
     pub require_affirmed_anchor_blocks: bool,
+    pub assume_present_anchor_blocks: bool,
     /// Fault injection for failing to push blocks
     pub fault_injection_block_push_fail_probability: Option<u8>,
     // fault injection for hiding blocks.
@@ -1939,6 +1940,7 @@ impl Default for NodeConfig {
             use_test_genesis_chainstate: None,
             always_use_affirmation_maps: true,
             require_affirmed_anchor_blocks: true,
+            assume_present_anchor_blocks: true,
             fault_injection_block_push_fail_probability: None,
             fault_injection_hide_blocks: false,
             chain_liveness_poll_time_secs: 300,
@@ -2428,6 +2430,7 @@ pub struct NodeConfigFile {
     pub use_test_genesis_chainstate: Option<bool>,
     pub always_use_affirmation_maps: Option<bool>,
     pub require_affirmed_anchor_blocks: Option<bool>,
+    pub assume_present_anchor_blocks: Option<bool>,
     /// At most, how often should the chain-liveness thread
     ///  wake up the chains-coordinator. Defaults to 300s (5 min).
     pub chain_liveness_poll_time_secs: Option<u64>,
@@ -2509,6 +2512,10 @@ impl NodeConfigFile {
             // miners should always try to mine, even if they don't have the anchored
             // blocks in the canonical affirmation map. Followers, however, can stall.
             require_affirmed_anchor_blocks: self.require_affirmed_anchor_blocks.unwrap_or(!miner),
+            // as of epoch 3.0, all prepare phases have anchor blocks.
+            // at the start of epoch 3.0, the chain stalls without anchor blocks.
+            // only set this to false if you're doing some very extreme testing.
+            assume_present_anchor_blocks: true,
             // chainstate fault_injection activation for hide_blocks.
             // you can't set this in the config file.
             fault_injection_hide_blocks: false,
diff --git a/testnet/stacks-node/src/run_loop/nakamoto.rs b/testnet/stacks-node/src/run_loop/nakamoto.rs
@@ -319,6 +319,7 @@ impl RunLoop {
                 let mut fee_estimator = moved_config.make_fee_estimator();
 
                 let coord_config = ChainsCoordinatorConfig {
+                    assume_present_anchor_blocks: moved_config.node.assume_present_anchor_blocks,
                     always_use_affirmation_maps: moved_config.node.always_use_affirmation_maps,
                     require_affirmed_anchor_blocks: moved_config
                         .node
diff --git a/testnet/stacks-node/src/run_loop/neon.rs b/testnet/stacks-node/src/run_loop/neon.rs
@@ -615,6 +615,7 @@ impl RunLoop {
                 let mut fee_estimator = moved_config.make_fee_estimator();
 
                 let coord_config = ChainsCoordinatorConfig {
+                    assume_present_anchor_blocks: moved_config.node.assume_present_anchor_blocks,
                     always_use_affirmation_maps: moved_config.node.always_use_affirmation_maps,
                     require_affirmed_anchor_blocks: moved_config
                         .node
@@ -1146,19 +1147,8 @@ impl RunLoop {
 
         let mut sortition_db_height = rc_aligned_height;
         let mut burnchain_height = sortition_db_height;
-        let mut num_sortitions_in_last_cycle = 1;
 
         // prepare to fetch the first reward cycle!
-        let mut target_burnchain_block_height = cmp::min(
-            burnchain_config.reward_cycle_to_block_height(
-                burnchain_config
-                    .block_height_to_reward_cycle(burnchain_height)
-                    .expect("BUG: block height is not in a reward cycle")
-                    + 1,
-            ),
-            burnchain.get_headers_height() - 1,
-        );
-
         debug!("Runloop: Begin main runloop starting a burnchain block {sortition_db_height}");
 
         let mut last_tenure_sortition_height = 0;
@@ -1186,17 +1176,13 @@ impl RunLoop {
 
             let remote_chain_height = burnchain.get_headers_height() - 1;
 
-            // wait for the p2p state-machine to do at least one pass
-            debug!("Runloop: Wait until Stacks block downloads reach a quiescent state before processing more burnchain blocks"; "remote_chain_height" => remote_chain_height, "local_chain_height" => burnchain_height);
-
-            // wait until it's okay to process the next reward cycle's sortitions
-            let ibd = match self.get_pox_watchdog().pox_sync_wait(
+            // wait until it's okay to process the next reward cycle's sortitions.
+            let (ibd, target_burnchain_block_height) = match self.get_pox_watchdog().pox_sync_wait(
                 &burnchain_config,
                 &burnchain_tip,
                 remote_chain_height,
-                num_sortitions_in_last_cycle,
             ) {
-                Ok(ibd) => ibd,
+                Ok(x) => x,
                 Err(e) => {
                     debug!("Runloop: PoX sync wait routine aborted: {e:?}");
                     continue;
@@ -1210,9 +1196,6 @@ impl RunLoop {
                 0.0
             };
 
-            // will recalculate this in the following loop
-            num_sortitions_in_last_cycle = 0;
-
             // Download each burnchain block and process their sortitions.  This, in turn, will
             // cause the node's p2p and relayer threads to go fetch and download Stacks blocks and
             // process them.  This loop runs for one reward cycle, so that the next pass of the
@@ -1260,8 +1243,6 @@ impl RunLoop {
                         "Runloop: New burnchain block height {next_sortition_height} > {sortition_db_height}"
                     );
 
-                    let mut sort_count = 0;
-
                     debug!("Runloop: block mining until we process all sortitions");
                     signal_mining_blocked(globals.get_miner_status());
 
@@ -1279,9 +1260,6 @@ impl RunLoop {
                                     "Failed to find block in fork processed by burnchain indexer",
                                 )
                         };
-                        if block.sortition {
-                            sort_count += 1;
-                        }
 
                         let sortition_id = &block.sortition_id;
 
@@ -1328,9 +1306,8 @@ impl RunLoop {
                     debug!("Runloop: enable miner after processing sortitions");
                     signal_mining_ready(globals.get_miner_status());
 
-                    num_sortitions_in_last_cycle = sort_count;
                     debug!(
-                        "Runloop: Synchronized sortitions up to block height {next_sortition_height} from {sortition_db_height} (chain tip height is {burnchain_height}); {num_sortitions_in_last_cycle} sortitions"
+                        "Runloop: Synchronized sortitions up to block height {next_sortition_height} from {sortition_db_height} (chain tip height is {burnchain_height})"
                     );
 
                     sortition_db_height = next_sortition_height;
@@ -1349,22 +1326,6 @@ impl RunLoop {
                 }
             }
 
-            // advance one reward cycle at a time.
-            // If we're still downloading, then this is simply target_burnchain_block_height + reward_cycle_len.
-            // Otherwise, this is burnchain_tip + reward_cycle_len
-            let next_target_burnchain_block_height = cmp::min(
-                burnchain_config.reward_cycle_to_block_height(
-                    burnchain_config
-                        .block_height_to_reward_cycle(target_burnchain_block_height)
-                        .expect("FATAL: burnchain height before system start")
-                        + 1,
-                ),
-                remote_chain_height,
-            );
-
-            debug!("Runloop: Advance target burnchain block height from {target_burnchain_block_height} to {next_target_burnchain_block_height} (sortition height {sortition_db_height})");
-            target_burnchain_block_height = next_target_burnchain_block_height;
-
             if sortition_db_height >= burnchain_height && !ibd {
                 let canonical_stacks_tip_height =
                     SortitionDB::get_canonical_burn_chain_tip(burnchain.sortdb_ref().conn())
diff --git a/testnet/stacks-node/src/syncctl.rs b/testnet/stacks-node/src/syncctl.rs
diff --git a/testnet/stacks-node/src/tests/nakamoto_integrations.rs b/testnet/stacks-node/src/tests/nakamoto_integrations.rs
