Rework call cache eviction to use ctid-based join deletes

graph/src/blockchain/mock.rs

@@ -6,7 +6,7 @@ use crate::{
         network_provider::ChainName,
         store::{
             BlockNumber, ChainHeadStore, ChainIdStore, DeploymentCursorTracker, DeploymentLocator,
-            SourceableStore,
+            SourceableStore, StaleCallCacheResult,
         },
         subgraph::InstanceDSTemplateInfo,
     },
@@ -595,8 +595,8 @@ impl ChainStore for MockChainStore {
     async fn clear_stale_call_cache(
         &self,
         _ttl_days: usize,
-        _ttl_max_contracts: Option<usize>,
-    ) -> Result<(), Error> {
+        _max_contracts: Option<usize>,
+    ) -> Result<StaleCallCacheResult, Error> {
         unimplemented!()
     }
     async fn chain_identifier(&self) -> Result<ChainIdentifier, Error> {

graph/src/components/store/traits.rs

@@ -530,6 +530,18 @@ pub trait ChainIdStore: Send + Sync + 'static {
 /// configuration can change this for individual chains
 pub const BLOCK_CACHE_SIZE: BlockNumber = i32::MAX;
 
+/// Result of clearing stale call cache entries.
+pub struct StaleCallCacheResult {
+    /// The effective TTL in days that was actually used for deletion.
+    /// This may be larger than the requested TTL if `max_contracts`
+    /// was set and caused the cutoff to be adjusted.
+    pub effective_ttl_days: usize,
+    /// Number of cache entries deleted from the call cache.
+    pub cache_entries_deleted: usize,
+    /// Number of contract entries deleted from call meta.
+    pub contracts_deleted: usize,
+}
+
 /// Common trait for blockchain store implementations.
 #[async_trait]
 pub trait ChainStore: ChainHeadStore {
@@ -657,12 +669,14 @@ pub trait ChainStore: ChainHeadStore {
     /// Clears call cache of the chain for the given `from` and `to` block number.
     async fn clear_call_cache(&self, from: BlockNumber, to: BlockNumber) -> Result<(), Error>;
 
-    /// Clears stale call cache entries for the given TTL in days.
+    /// Clears stale call cache entries for the given TTL in days. If
+    /// `max_contracts` is set, increase the effective TTL so that at
+    /// most `max_contracts` contracts are evicted.
     async fn clear_stale_call_cache(
         &self,
         ttl_days: usize,
-        ttl_max_contracts: Option<usize>,
-    ) -> Result<(), Error>;
+        max_contracts: Option<usize>,
+    ) -> Result<StaleCallCacheResult, Error>;
 
     /// Return the chain identifier for this store.
     async fn chain_identifier(&self) -> Result<ChainIdentifier, Error>;

graph/src/env/store.rs

@@ -171,11 +171,6 @@ pub struct EnvVarsStore {
     /// Disables storing or reading `eth_call` results from the store call cache.
     /// Set by `GRAPH_STORE_DISABLE_CALL_CACHE`. Defaults to false.
     pub disable_call_cache: bool,
-    /// The number of contracts to delete from the call cache in one batch
-    /// when clearing stale entries, set by
-    /// `GRAPH_STORE_STALE_CALL_CACHE_CONTRACTS_BATCH_SIZE`. The default
-    /// value is 100 contracts.
-    pub stale_call_cache_contracts_batch_size: usize,
     /// Set by `GRAPH_STORE_DISABLE_CHAIN_HEAD_PTR_CACHE`. Default is false.
     /// Set to true to disable chain_head_ptr caching (safety escape hatch).
     pub disable_chain_head_ptr_cache: bool,
@@ -253,7 +248,6 @@ impl TryFrom<InnerStore> for EnvVarsStore {
             account_like_min_versions_count: x.account_like_min_versions_count,
             account_like_max_unique_ratio: x.account_like_max_unique_ratio.map(|r| r.0),
             disable_call_cache: x.disable_call_cache,
-            stale_call_cache_contracts_batch_size: x.stale_call_cache_contracts_batch_size,
             disable_chain_head_ptr_cache: x.disable_chain_head_ptr_cache,
             connection_validation_idle_secs: Duration::from_secs(x.connection_validation_idle_secs),
             connection_unavailable_retry: Duration::from_secs(
@@ -370,11 +364,6 @@ pub struct InnerStore {
     account_like_max_unique_ratio: Option<ZeroToOneF64>,
     #[envconfig(from = "GRAPH_STORE_DISABLE_CALL_CACHE", default = "false")]
     disable_call_cache: bool,
-    #[envconfig(
-        from = "GRAPH_STORE_STALE_CALL_CACHE_CONTRACTS_BATCH_SIZE",
-        default = "100"
-    )]
-    stale_call_cache_contracts_batch_size: usize,
     #[envconfig(from = "GRAPH_STORE_DISABLE_CHAIN_HEAD_PTR_CACHE", default = "false")]
     disable_chain_head_ptr_cache: bool,
     #[envconfig(from = "GRAPH_STORE_CONNECTION_VALIDATION_IDLE_SECS", default = "30")]

graph/src/lib.rs

@@ -128,8 +128,8 @@ pub mod prelude {
         EntityCollection, EntityFilter, EntityLink, EntityOperation, EntityOrder,
         EntityOrderByChild, EntityOrderByChildInfo, EntityQuery, EntityRange, EntityWindow,
         EthereumCallCache, ParentLink, PartialBlockPtr, PoolWaitStats, QueryStore,
-        QueryStoreManager, SeqGenerator, StoreError, StoreEvent, StoreEventStreamBox,
-        SubgraphStore, UnfailOutcome, WindowAttribute, BLOCK_NUMBER_MAX,
+        QueryStoreManager, SeqGenerator, StaleCallCacheResult, StoreError, StoreEvent,
+        StoreEventStreamBox, SubgraphStore, UnfailOutcome, WindowAttribute, BLOCK_NUMBER_MAX,
     };
     pub use crate::components::subgraph::{
         BlockState, BlockStateCheckpoint, HostMetrics, InstanceDSTemplateInfo, RuntimeHost,

node/src/bin/manager.rs

@@ -620,10 +620,11 @@ pub enum CallCacheCommand {
         remove_entire_cache: bool,
         /// Remove the cache for contracts that have not been accessed in the last <TTL_DAYS> days
         #[clap(long, conflicts_with_all = &["from", "to", "remove-entire-cache"], value_parser = clap::value_parser!(u32).range(1..))]
-        ttl_days: Option<usize>,
-        /// Limits the number of contracts to consider for cache removal when using --ttl_days
-        #[clap(long, conflicts_with_all = &["remove-entire-cache", "to", "from"], requires = "ttl_days", value_parser = clap::value_parser!(u64).range(1..))]
-        ttl_max_contracts: Option<usize>,
+        ttl_days: Option<u32>,
+        /// Maximum number of contracts to evict. When set, the effective TTL
+        /// is increased so that at most this many contracts are deleted.
+        #[clap(long, requires = "ttl_days")]
+        max_contracts: Option<usize>,
         /// Starting block number
         #[clap(long, short, conflicts_with = "remove-entire-cache", requires = "to")]
         from: Option<i32>,
@@ -1549,14 +1550,14 @@ async fn main() -> anyhow::Result<()> {
                             to,
                             remove_entire_cache,
                             ttl_days,
-                            ttl_max_contracts,
+                            max_contracts,
                         } => {
                             let chain_store = ctx.chain_store(&chain_name).await?;
                             if let Some(ttl_days) = ttl_days {
                                 return commands::chain::clear_stale_call_cache(
                                     chain_store,
-                                    ttl_days,
-                                    ttl_max_contracts,
+                                    ttl_days as usize,
+                                    max_contracts,
                                 )
                                 .await;
                             }

node/src/manager/commands/chain.rs

@@ -84,15 +84,27 @@ pub async fn clear_call_cache(
 pub async fn clear_stale_call_cache(
     chain_store: Arc<ChainStore>,
     ttl_days: usize,
-    ttl_max_contracts: Option<usize>,
+    max_contracts: Option<usize>,
 ) -> Result<(), Error> {
     println!(
         "Removing stale entries from the call cache for `{}`",
         chain_store.chain
     );
-    chain_store
-        .clear_stale_call_cache(ttl_days, ttl_max_contracts)
+    let result = chain_store
+        .clear_stale_call_cache(ttl_days, max_contracts)
         .await?;
+    if result.effective_ttl_days != ttl_days {
+        println!(
+            "Effective TTL: {} days (adjusted from {} to stay within {} contracts)",
+            result.effective_ttl_days,
+            ttl_days,
+            max_contracts.unwrap()
+        );
+    }
+    println!(
+        "Deleted {} cache entries for {} contracts",
+        result.cache_entries_deleted, result.contracts_deleted
+    );
     Ok(())
 }