store: Limit for how many prune runs we keep status info

Author: lutter

graph/src/env/store.rs

@@ -111,6 +111,9 @@ pub struct EnvVarsStore {
     /// blocks) than its history limit. The default value is 1.2 and the
     /// value must be at least 1.01
     pub history_slack_factor: f64,
+    /// For how many prune runs per deployment to keep status information.
+    /// Set by `GRAPH_STORE_HISTORY_KEEP_STATUS`. The default is 5
+    pub prune_keep_history: usize,
     /// How long to accumulate changes into a batch before a write has to
     /// happen. Set by the environment variable
     /// `GRAPH_STORE_WRITE_BATCH_DURATION` in seconds. The default is 300s.
@@ -184,6 +187,7 @@ impl TryFrom<InnerStore> for EnvVarsStore {
             rebuild_threshold: x.rebuild_threshold.0,
             delete_threshold: x.delete_threshold.0,
             history_slack_factor: x.history_slack_factor.0,
+            prune_keep_history: x.prune_keep_status,
             write_batch_duration: Duration::from_secs(x.write_batch_duration_in_secs),
             write_batch_size: x.write_batch_size * 1_000,
             create_gin_indexes: x.create_gin_indexes,
@@ -257,6 +261,8 @@ pub struct InnerStore {
     delete_threshold: ZeroToOneF64,
     #[envconfig(from = "GRAPH_STORE_HISTORY_SLACK_FACTOR", default = "1.2")]
     history_slack_factor: HistorySlackF64,
+    #[envconfig(from = "GRAPH_STORE_HISTORY_KEEP_STATUS", default = "5")]
+    prune_keep_status: usize,
     #[envconfig(from = "GRAPH_STORE_WRITE_BATCH_DURATION", default = "300")]
     write_batch_duration_in_secs: u64,
     #[envconfig(from = "GRAPH_STORE_WRITE_BATCH_SIZE", default = "10000")]

node/src/manager/commands/prune.rs

@@ -341,14 +341,16 @@ pub async fn status(
         let runs = match runs.len() {
             0 => unreachable!("we checked that runs is not empty"),
             1 => format!("There is only one prune run #{}", runs[0]),
+            2 => format!("Only prune runs #{} and #{} exist", runs[0], runs[1]),
             _ => format!(
-                "Only prune runs #{} up to #{} exist",
+                "Only prune runs #{} and #{} up to #{} exist",
                 runs[0],
+                runs[1],
                 runs.last().unwrap()
             ),
         };
         return Err(anyhow!(
-            "No information about prune run #{run} found for deployment {deployment}. {runs}"
+            "No information about prune run #{run} found for deployment {deployment}.\n  {runs}"
         ));
     };
     println!("prune {deployment} (run #{run})");

store/postgres/src/relational/prune.rs

@@ -508,6 +508,7 @@ mod status {
     };
     use graph::{
         components::store::{PruneRequest, PruningStrategy, StoreResult},
+        env::ENV_VARS,
         prelude::StoreError,
     };
 
@@ -676,6 +677,15 @@ mod status {
                 .unwrap_or(0)
                 + 1;
 
+            // Delete old prune state. Keep the initial run and the last
+            // `prune_keep_history` runs (including this one)
+            diesel::delete(ps::table)
+                .filter(ps::id.eq(layout.site.id))
+                .filter(ps::run.gt(1))
+                .filter(ps::run.lt(run - (ENV_VARS.store.prune_keep_history - 1) as i32))
+                .execute(conn)
+                .map_err(StoreError::from)?;
+
             Ok(Tracker { layout, run })
         }