Merge pull request #30 from graphprotocol/tmigone/deployment-cache

feat: implement deployment cache to reduce IPFS request load

availability-oracle/src/main.rs

@@ -18,6 +18,7 @@ use manifest::{Abi, DataSource, Manifest, Mapping};
 use network_subgraph::*;
 use secp256k1::SecretKey;
 use std::sync::Arc;
+use std::time::SystemTime;
 use std::time::{Duration, Instant};
 use std::{fmt::Display, str::FromStr};
 use structopt::StructOpt;
@@ -148,6 +149,8 @@ struct Config {
     pub oracle_index: Option<u64>,
 }
 
+const VALID_DEPLOYMENT_CACHE_TTL: Duration = Duration::from_secs(60 * 60 * 24);
+
 #[tokio::main]
 async fn main() -> Result<()> {
     common::main(run).await
@@ -159,6 +162,10 @@ async fn run(logger: Logger, config: Config) -> Result<()> {
     let epoch_subgraph =
         EpochBlockOracleSubgraphImpl::new(logger.clone(), config.epoch_block_oracle_subgraph);
     let contract: Box<dyn StateManager> = if config.dry_run {
+        info!(
+            logger,
+            "Running in dry mode: no transactions will be submitted on chain!"
+        );
         Box::new(StateManagerDryRun::new(logger.clone()))
     } else {
         let signing_key: &SecretKey = &config.signing_key.unwrap().parse()?;
@@ -181,6 +188,10 @@ async fn run(logger: Logger, config: Config) -> Result<()> {
     if config.period > Duration::from_secs(0) {
         let mut interval = tokio::time::interval(config.period);
         interval.set_missed_tick_behavior(MissedTickBehavior::Skip);
+
+        // Valid deployments get checked only every VALID_DEPLOYMENT_CACHE_TTL seconds
+        let mut valid_deployment_cache: Vec<(Cid, SystemTime)> = Vec::new();
+
         loop {
             interval.tick().await;
 
@@ -197,11 +208,16 @@ async fn run(logger: Logger, config: Config) -> Result<()> {
                 grace_period,
                 epoch_subgraph.clone(),
                 &config.supported_data_source_kinds,
+                valid_deployment_cache.clone(),
             )
             .await
             {
-                Ok(()) => {
+                Ok(updated_deployment_cache) => {
                     METRICS.reconcile_runs_ok.inc();
+                    valid_deployment_cache = updated_deployment_cache;
+                    info!(logger, "Deployment cache updated";
+                        "count" => valid_deployment_cache.len()
+                    );
                 }
                 Err(e) => {
                     METRICS.reconcile_runs_err.inc();
@@ -221,7 +237,7 @@ async fn run(logger: Logger, config: Config) -> Result<()> {
             ipfs.invalidate_cache();
         }
     }
-    reconcile_deny_list(
+    match reconcile_deny_list(
         &logger,
         &ipfs,
         &*contract,
@@ -230,8 +246,13 @@ async fn run(logger: Logger, config: Config) -> Result<()> {
         grace_period,
         epoch_subgraph.clone(),
         &config.supported_data_source_kinds,
+        Vec::new(),
     )
     .await
+    {
+        Ok(_) => return Ok(()),
+        Err(e) => return Err(e),
+    }
 }
 
 // This function is used to create a state manager based on the configuration.
@@ -281,7 +302,8 @@ pub async fn reconcile_deny_list(
     grace_period: Duration,
     epoch_subgraph: Arc<impl EpochBlockOracleSubgraph>,
     supported_ds_kinds: &[String],
-) -> Result<(), Error> {
+    valid_deployment_cache: Vec<(Cid, SystemTime)>,
+) -> Result<Vec<(Cid, SystemTime)>, Error> {
     let logger = logger.clone();
 
     // Fetch supported networks
@@ -300,20 +322,35 @@ pub async fn reconcile_deny_list(
     );
 
     // Check the availability status of all subgraphs, and gather which should flip the deny flag.
-    let status_changes: Vec<([u8; 32], bool)> = subgraph
+    let deployment_status: Vec<([u8; 32], bool, bool, SystemTime)> = subgraph
         .deployments_over_threshold(min_signal, grace_period)
         .map(|deployment| async {
             let deployment = deployment?;
             let id = bytes32_to_cid_v0(deployment.id);
-            let validity = match check(ipfs, id, &supported_networks, supported_ds_kinds).await {
-                Ok(()) => Valid::Yes,
-                Err(CheckError::Invalid(e)) => Valid::No(e),
-                Err(CheckError::Other(e)) => return Err(e),
+
+            // Valid subgraphs are only checked every VALID_DEPLOYMENT_CACHE_TTL seconds to reduce IPFS requests
+            let cached = valid_deployment_cache
+                .iter()
+                .filter(|(_, last_validated)| {
+                    last_validated.elapsed().unwrap() < VALID_DEPLOYMENT_CACHE_TTL
+                })
+                .find(|(cid, _)| *cid == id);
+
+            if cached.is_some() {
+                METRICS.valid_deployment_cache_hits.inc();
+                return Ok((deployment, Valid::Yes, cached.unwrap().1));
+            } else {
+                let validity = match check(ipfs, id, &supported_networks, supported_ds_kinds).await
+                {
+                    Ok(()) => Valid::Yes,
+                    Err(CheckError::Invalid(e)) => Valid::No(e),
+                    Err(CheckError::Other(e)) => return Err(e),
+                };
+                return Ok((deployment, validity, SystemTime::now()));
             };
-            Result::<_, Error>::Ok((deployment, validity))
         })
         .buffered(100)
-        .try_filter_map(move |(deployment, validity)| {
+        .try_filter_map(move |(deployment, validity, last_validated)| {
             let logger = logger.clone();
             async move {
                 info!(logger, "Check subgraph";
@@ -336,7 +373,7 @@ pub async fn reconcile_deny_list(
                                 );
                             }
                         };
-                        None
+                        Some((deployment.id, should_deny, false, last_validated))
                     }
 
                     // The validity status changed, flip the deny flag.
@@ -347,15 +384,32 @@ pub async fn reconcile_deny_list(
                                         "status" => should_deny,
                                         "reason" => validity.to_string(),
                         );
-                        Some((deployment.id, should_deny))
+                        Some((deployment.id, should_deny, true, last_validated))
                     }
                 })
             }
         })
         .try_collect()
         .await?;
 
-    state_manager.deny_many(status_changes).await
+    // Flip on chain status for those deployments that changed
+    let changed_deployments = deployment_status
+        .iter()
+        .filter(|(_, _, status_changed, _)| *status_changed)
+        .map(|(cid, should_deny, _, _)| (*cid, *should_deny))
+        .collect();
+    match state_manager.deny_many(changed_deployments).await {
+        Ok(_) => {}
+        Err(e) => return Err(e),
+    };
+
+    // Return updated deployment cache
+    let updated_deployment_cache: Vec<(Cid, SystemTime)> = deployment_status
+        .iter()
+        .filter(|(_, should_deny, _, _)| !*should_deny)
+        .map(|(cid, _, _, last_validated)| (bytes32_to_cid_v0(*cid), *last_validated))
+        .collect();
+    Ok(updated_deployment_cache)
 }
 
 enum Valid {
@@ -537,6 +591,7 @@ struct Metrics {
     reconcile_runs_total: prometheus::IntCounter,
     reconcile_runs_ok: prometheus::IntCounter,
     reconcile_runs_err: prometheus::IntCounter,
+    valid_deployment_cache_hits: prometheus::IntCounter,
 }
 
 lazy_static! {
@@ -561,6 +616,11 @@ impl Metrics {
                 "Total reconcile runs with errors"
             )
             .unwrap(),
+            valid_deployment_cache_hits: prometheus::register_int_counter!(
+                "valid_deployment_cache_hits",
+                "Total valid deployment cache hits"
+            )
+            .unwrap(),
         }
     }
 }

availability-oracle/src/test.rs

@@ -61,9 +61,10 @@ mod tests {
                 "file/ipfs".into(),
                 "substreams".into(),
             ],
+            vec![],
         )
         .await
-        .unwrap()
+        .unwrap();
     }
 
     struct MockSubgraph;