Merge of #4245

.changelog/unreleased/bug-fixes/4245-fix-tm-mode.md

@@ -0,0 +1,5 @@
+- Fixed running CometBFT as a validator when the Namada config `tendermint_mode`
+  is set to a non-validator mode. When the `tendermint_mode` changes
+  from a validator to non-validator mode, the node will replace and
+  backup the validator consensus key and state in the CometBFT directory.
+  ([\#4245](https://github.com/anoma/namada/pull/4245))

crates/apps/src/bin/namada-node/cli.rs

@@ -28,19 +28,15 @@ pub fn main() -> Result<()> {
                     );
                     ScheduledMigration::from_path(p, hash, height).unwrap()
                 });
-                node::run(
-                    chain_ctx.config.ledger,
-                    wasm_dir,
-                    scheduled_migration,
-                );
+                node::run(chain_ctx.config, wasm_dir, scheduled_migration);
             }
             cmds::Ledger::RunUntil(cmds::LedgerRunUntil(args)) => {
                 let mut chain_ctx = ctx.take_chain_or_exit();
                 let wasm_dir = chain_ctx.wasm_dir();
                 sleep_until(args.time);
                 chain_ctx.config.ledger.shell.action_at_height =
                     Some(args.action_at_height);
-                node::run(chain_ctx.config.ledger, wasm_dir, None);
+                node::run(chain_ctx.config, wasm_dir, None);
             }
             cmds::Ledger::Reset(_) => {
                 let chain_ctx = ctx.take_chain_or_exit();

crates/apps_lib/src/config/mod.rs

@@ -51,7 +51,7 @@ pub struct NodeLocalConfig {
     pub recheck_process_proposal: bool,
 }
 
-#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq)]
+#[derive(Clone, Copy, Debug, Serialize, Deserialize, PartialEq, Eq)]
 pub enum TendermintMode {
     Full,
     Validator,
@@ -101,7 +101,6 @@ pub struct Ledger {
 #[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct Shell {
     pub base_dir: PathBuf,
-    // pub ledger_address: SocketAddr,
     /// RocksDB block cache maximum size in bytes.
     /// When not set, defaults to 1/3 of the available memory.
     pub block_cache_bytes: Option<u64>,
@@ -122,6 +121,8 @@ pub struct Shell {
     pub action_at_height: Option<ActionAtHeight>,
     /// Specify if tendermint is started as validator, fullnode or seednode
     pub tendermint_mode: TendermintMode,
+    /// A `tendermint_mode` set on the last node start-up, if any.
+    pub last_tendermint_mode: Option<TendermintMode>,
     /// When set, indicates after how many blocks a new snapshot
     /// will be taken (counting from the first block)
     pub blocks_between_snapshots: Option<NonZeroU64>,
@@ -155,6 +156,7 @@ impl Ledger {
                 cometbft_dir: COMETBFT_DIR.into(),
                 action_at_height: None,
                 tendermint_mode: mode,
+                last_tendermint_mode: None,
                 blocks_between_snapshots: None,
                 snapshots_to_keep: None,
             },

crates/apps_lib/src/tendermint_node.rs

@@ -1,7 +1,9 @@
+use std::fs;
 use std::path::{Path, PathBuf};
 
 use namada_sdk::borsh::BorshSerializeExt;
 use namada_sdk::key::*;
+use rand_core::OsRng;
 use serde_json::json;
 use sha2::{Digest, Sha256};
 use thiserror::Error;
@@ -98,6 +100,32 @@ pub fn write_validator_state(home_dir: impl AsRef<Path>) -> Result<()> {
     write_validator(validator_state(home_dir), STATE_DIR, STATE_FILE, state)
 }
 
+/// Move the current validator key and state into a backup file (adds ".bk"
+/// extension), if any.
+pub fn backup_validator_key_and_state(home_dir: impl AsRef<Path>) {
+    let key_path = validator_key(&home_dir);
+    let state_path = validator_state(&home_dir);
+    for path in [key_path, state_path] {
+        if path.exists() {
+            fs::rename(&path, path.with_extension("json.bk")).expect(
+                "Must be able to backup existing CometBFT validator key and \
+                 state",
+            );
+        }
+    }
+}
+
+/// Write a dummy validator key and empty state file.
+pub fn write_dummy_validator_key_and_state(home_dir: impl AsRef<Path>) {
+    let dummy_key: common::SecretKey = ed25519::SigScheme::generate(&mut OsRng)
+        .try_to_sk()
+        .unwrap();
+    write_validator_key(&home_dir, &dummy_key)
+        .expect("Must be able to write dummy validator key.");
+    write_validator_state(&home_dir)
+        .expect("Must be able to write dummy validator state.");
+}
+
 /// Abstract over the initialization of validator data for Tendermint
 pub fn write_validator(
     path: PathBuf,
@@ -107,7 +135,7 @@ pub fn write_validator(
 ) -> Result<()> {
     let parent_dir = path.parent().unwrap();
     // Make sure the dir exists
-    std::fs::create_dir_all(parent_dir).map_err(|err| {
+    fs::create_dir_all(parent_dir).map_err(|err| {
         Error::CantCreate(format!(
             "{} at {}. Caused by {err}",
             err_dir,
@@ -152,8 +180,8 @@ pub fn id_from_pk(pk: &common::PublicKey) -> TendermintNodeId {
     TendermintNodeId::new(bytes)
 }
 
-fn ensure_empty(path: &PathBuf) -> std::io::Result<std::fs::File> {
-    std::fs::OpenOptions::new()
+fn ensure_empty(path: &PathBuf) -> std::io::Result<fs::File> {
+    fs::OpenOptions::new()
         .create(true)
         .write(true)
         .truncate(true)

crates/node/src/lib.rs

@@ -284,10 +284,12 @@ fn emit_warning_on_non_64bit_cpu() {
 
 /// Run the ledger with an async runtime
 pub fn run(
-    config: config::Ledger,
+    config: config::Config,
     wasm_dir: PathBuf,
     scheduled_migration: Option<ScheduledMigration>,
 ) {
+    handle_tendermint_mode_change(&config);
+
     emit_warning_on_non_64bit_cpu();
 
     let logical_cores = num_cpus::get();
@@ -323,7 +325,54 @@ pub fn run(
         .enable_all()
         .build()
         .unwrap()
-        .block_on(run_aux(config, wasm_dir, scheduled_migration));
+        .block_on(run_aux(config.ledger, wasm_dir, scheduled_migration));
+}
+
+/// Check the `tendermint_mode` has changed from validator to non-validator
+/// mode, in which case we replace and backup the validator keys and state to
+/// avoid CometBFT running as a validator. We also persist the
+/// `last_tendermint_node` in the config for the next run.
+fn handle_tendermint_mode_change(config: &config::Config) {
+    // Check if the node was previously ran as a Validator, but isn't anymore
+    if !matches!(
+        config.ledger.shell.tendermint_mode,
+        TendermintMode::Validator
+    ) && matches!(
+        config.ledger.shell.last_tendermint_mode,
+        Some(TendermintMode::Validator)
+    ) {
+        // Backup and replace CometBFT validator key and state
+        let cometbft_dir = config.ledger.cometbft_dir();
+        namada_apps_lib::tendermint_node::backup_validator_key_and_state(
+            &cometbft_dir,
+        );
+        namada_apps_lib::tendermint_node::write_dummy_validator_key_and_state(
+            &cometbft_dir,
+        );
+    }
+
+    if config.ledger.shell.last_tendermint_mode.is_none()
+        || config.ledger.shell.last_tendermint_mode
+            != Some(config.ledger.shell.tendermint_mode)
+    {
+        let mut config = config.clone();
+        config.ledger.shell.last_tendermint_mode =
+            Some(config.ledger.shell.tendermint_mode);
+        // Remove this field in case it's set from running `ledger run-until` -
+        // it shouldn't be persisted
+        config.ledger.shell.action_at_height = None;
+        let replace = true;
+        config
+            .write(
+                &config.ledger.shell.base_dir,
+                &config.ledger.chain_id,
+                replace,
+            )
+            .expect(
+                "Must be able to persist config with changed \
+                 `last_tendermint_mode`.",
+            );
+    }
 }
 
 /// Resets the tendermint_node state and removes database files
@@ -600,7 +649,7 @@ fn start_abci_broadcaster_shell(
     );
 
     // Construct our ABCI application.
-    let tendermint_mode = config.shell.tendermint_mode.clone();
+    let tendermint_mode = config.shell.tendermint_mode;
     let proxy_app_address =
         convert_tm_addr_to_socket_addr(&config.cometbft.proxy_app);