Override neon.max_cluster_size for the time of compute_ctl (#5998)

Temporarily reset neon.max_cluster_size to avoid
the possibility of hitting the limit, while we are applying config:
creating new extensions, roles, etc...

compute_tools/src/compute.rs

@@ -728,7 +728,12 @@ impl ComputeNode {
 
         // Write new config
         let pgdata_path = Path::new(&self.pgdata);
-        config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), &spec, None)?;
+        let postgresql_conf_path = pgdata_path.join("postgresql.conf");
+        config::write_postgres_conf(&postgresql_conf_path, &spec, None)?;
+        // temporarily reset max_cluster_size in config
+        // to avoid the possibility of hitting the limit, while we are reconfiguring:
+        // creating new extensions, roles, etc...
+        config::compute_ctl_temp_override_create(pgdata_path, "neon.max_cluster_size=-1")?;
         self.pg_reload_conf()?;
 
         let mut client = Client::connect(self.connstr.as_str(), NoTls)?;
@@ -749,6 +754,10 @@ impl ComputeNode {
         // 'Close' connection
         drop(client);
 
+        // reset max_cluster_size in config back to original value and reload config
+        config::compute_ctl_temp_override_remove(pgdata_path)?;
+        self.pg_reload_conf()?;
+
         let unknown_op = "unknown".to_string();
         let op_id = spec.operation_uuid.as_ref().unwrap_or(&unknown_op);
         info!(
@@ -809,7 +818,17 @@ impl ComputeNode {
 
         let config_time = Utc::now();
         if pspec.spec.mode == ComputeMode::Primary && !pspec.spec.skip_pg_catalog_updates {
+            let pgdata_path = Path::new(&self.pgdata);
+            // temporarily reset max_cluster_size in config
+            // to avoid the possibility of hitting the limit, while we are applying config:
+            // creating new extensions, roles, etc...
+            config::compute_ctl_temp_override_create(pgdata_path, "neon.max_cluster_size=-1")?;
+            self.pg_reload_conf()?;
+
             self.apply_config(&compute_state)?;
+
+            config::compute_ctl_temp_override_remove(pgdata_path)?;
+            self.pg_reload_conf()?;
         }
 
         let startup_end_time = Utc::now();

compute_tools/src/config.rs

@@ -93,5 +93,25 @@ pub fn write_postgres_conf(
         writeln!(file, "neon.extension_server_port={}", port)?;
     }
 
+    // This is essential to keep this line at the end of the file,
+    // because it is intended to override any settings above.
+    writeln!(file, "include_if_exists = 'compute_ctl_temp_override.conf'")?;
+
+    Ok(())
+}
+
+/// create file compute_ctl_temp_override.conf in pgdata_dir
+/// add provided options to this file
+pub fn compute_ctl_temp_override_create(pgdata_path: &Path, options: &str) -> Result<()> {
+    let path = pgdata_path.join("compute_ctl_temp_override.conf");
+    let mut file = File::create(path)?;
+    write!(file, "{}", options)?;
+    Ok(())
+}
+
+/// remove file compute_ctl_temp_override.conf in pgdata_dir
+pub fn compute_ctl_temp_override_remove(pgdata_path: &Path) -> Result<()> {
+    let path = pgdata_path.join("compute_ctl_temp_override.conf");
+    std::fs::remove_file(path)?;
     Ok(())
 }

compute_tools/src/spec.rs

@@ -118,19 +118,6 @@ pub fn get_spec_from_control_plane(
     spec
 }
 
-/// It takes cluster specification and does the following:
-/// - Serialize cluster config and put it into `postgresql.conf` completely rewriting the file.
-/// - Update `pg_hba.conf` to allow external connections.
-pub fn handle_configuration(spec: &ComputeSpec, pgdata_path: &Path) -> Result<()> {
-    // File `postgresql.conf` is no longer included into `basebackup`, so just
-    // always write all config into it creating new file.
-    config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), spec, None)?;
-
-    update_pg_hba(pgdata_path)?;
-
-    Ok(())
-}
-
 /// Check `pg_hba.conf` and update if needed to allow external connections.
 pub fn update_pg_hba(pgdata_path: &Path) -> Result<()> {
     // XXX: consider making it a part of spec.json

test_runner/regress/test_timeline_size.py

@@ -146,6 +146,72 @@ def wait_for_pageserver_catchup(endpoint_main: Endpoint, polling_interval=1, tim
         time.sleep(polling_interval)
 
 
+def test_timeline_size_quota_on_startup(neon_env_builder: NeonEnvBuilder):
+    env = neon_env_builder.init_start()
+    client = env.pageserver.http_client()
+    new_timeline_id = env.neon_cli.create_branch("test_timeline_size_quota_on_startup")
+
+    wait_for_timeline_size_init(client, tenant=env.initial_tenant, timeline=new_timeline_id)
+
+    endpoint_main = env.endpoints.create(
+        "test_timeline_size_quota_on_startup",
+        # Set small limit for the test
+        config_lines=["neon.max_cluster_size=30MB"],
+    )
+    endpoint_main.start()
+
+    log.info("postgres is running on 'test_timeline_size_quota_on_startup' branch")
+
+    with closing(endpoint_main.connect()) as conn:
+        with conn.cursor() as cur:
+            cur.execute("CREATE TABLE foo (t text)")
+
+            # Insert many rows. This query must fail because of space limit
+            try:
+                for _i in range(5000):
+                    cur.execute(
+                        """
+                        INSERT INTO foo
+                            SELECT 'long string to consume some space' || g
+                            FROM generate_series(1, 100) g
+                    """
+                    )
+
+                # If we get here, the timeline size limit failed
+                log.error("Query unexpectedly succeeded")
+                raise AssertionError()
+
+            except psycopg2.errors.DiskFull as err:
+                log.info(f"Query expectedly failed with: {err}")
+
+    # Restart endpoint that reached the limit to ensure that it doesn't fail on startup
+    # i.e. the size limit is not enforced during startup.
+    endpoint_main.stop()
+    # don't skip pg_catalog updates - it runs CREATE EXTENSION neon
+    # which is needed for neon.pg_cluster_size() to work
+    endpoint_main.respec(skip_pg_catalog_updates=False)
+    endpoint_main.start()
+
+    # ensure that the limit is enforced after startup
+    with closing(endpoint_main.connect()) as conn:
+        with conn.cursor() as cur:
+            # This query must fail because of space limit
+            try:
+                cur.execute(
+                    """
+                    INSERT INTO foo
+                        SELECT 'long string to consume some space' || g
+                        FROM generate_series(1, 100000) g
+                """
+                )
+                # If we get here, the timeline size limit failed
+                log.error("Query unexpectedly succeeded")
+                raise AssertionError()
+
+            except psycopg2.errors.DiskFull as err:
+                log.info(f"Query expectedly failed with: {err}")
+
+
 def test_timeline_size_quota(neon_env_builder: NeonEnvBuilder):
     env = neon_env_builder.init_start()
     client = env.pageserver.http_client()