Another no-op helm release

Revert "Do not unban replicas if a primary is available (#843)"

This reverts commit cdcfa99fb9.

.gitignore

@@ -12,3 +12,4 @@ dev/cache
 !dev/cache/.keepme
 .venv
 **/__pycache__
+.bundle

CONFIG.md

@@ -36,10 +36,11 @@ Port at which prometheus exporter listens on.
 ### connect_timeout
 ```
 path: general.connect_timeout
-default: 5000 # milliseconds
+default: 1000 # milliseconds
 ```
 
-How long to wait before aborting a server connection (ms).
+How long the client waits to obtain a server connection before aborting (ms).
+This is similar to PgBouncer's `query_wait_timeout`.
 
 ### idle_timeout
 ```
@@ -462,10 +463,18 @@ path: pools.<pool_name>.users.<user_index>.pool_size
 default: 9
 ```
 
-Maximum number of server connections that can be established for this user
+Maximum number of server connections that can be established for this user.
 The maximum number of connection from a single Pgcat process to any database in the cluster
 is the sum of pool_size across all users.
 
+### min_pool_size
+```
+path: pools.<pool_name>.users.<user_index>.min_pool_size
+default: 0
+```
+
+Minimum number of idle server connections to retain for this pool.
+
 ### statement_timeout
 ```
 path: pools.<pool_name>.users.<user_index>.statement_timeout
@@ -475,6 +484,16 @@ default: 0
 Maximum query duration. Dangerous, but protects against DBs that died in a non-obvious way.
 0 means it is disabled.
 
+### connect_timeout
+```
+path: pools.<pool_name>.users.<user_index>.connect_timeout
+default: <UNSET> # milliseconds
+```
+
+How long the client waits to obtain a server connection before aborting (ms).
+This is similar to PgBouncer's `query_wait_timeout`.
+If unset, uses the `connect_timeout` defined globally.
+
 ## `pools.<pool_name>.shards.<shard_index>` Section
 
 ### servers
@@ -502,4 +521,3 @@ default: "shard0"
 ```
 
 Database name (e.g. "postgres")
-

Cargo.lock

@@ -192,12 +192,11 @@ checksum = "604178f6c5c21f02dc555784810edfb88d34ac2c73b2eae109655649ee73ce3d"
 
 [[package]]
 name = "bb8"
-version = "0.8.1"
+version = "0.8.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "98b4b0f25f18bcdc3ac72bdb486ed0acf7e185221fd4dc985bc15db5800b0ba2"
+checksum = "d89aabfae550a5c44b43ab941844ffcd2e993cb6900b342debf59e9ea74acdb8"
 dependencies = [
  "async-trait",
- "futures-channel",
  "futures-util",
  "parking_lot",
  "tokio",

Cargo.toml

@@ -8,7 +8,7 @@ edition = "2021"
 tokio = { version = "1", features = ["full"] }
 bytes = "1"
 md-5 = "0.10"
-bb8 = "0.8.1"
+bb8 = "=0.8.6"
 async-trait = "0.1"
 rand = "0.8"
 chrono = "0.4"

charts/pgcat/Chart.yaml

@@ -5,4 +5,4 @@ maintainers:
   - name: Wildcard
     email: support@w6d.io
 appVersion: "1.2.0"
-version: 0.2.1
+version: 0.2.4

cr.yaml

@@ -1 +1,2 @@
 sign: false
+pages_branch: main

pgcat.toml

@@ -179,7 +179,7 @@ primary_reads_enabled = true
 # `random`: picks a shard at random
 # `random_healthy`: picks a shard at random favoring shards with the least number of recent errors
 # `shard_<number>`: e.g. shard_0, shard_4, etc. picks a specific shard, everytime
-# no_shard_specified_behavior = "shard_0"
+# default_shard = "shard_0"
 
 # So what if you wanted to implement a different hashing function,
 # or you've already built one and you want this pooler to use it?

src/client.rs

@@ -881,6 +881,7 @@ where
         };
 
         query_router.update_pool_settings(&pool.settings);
+        query_router.set_default_role();
 
         // Our custom protocol loop.
         // We expect the client to either start a transaction with regular queries

src/prometheus.rs

@@ -309,6 +309,7 @@ async fn prometheus_stats(
             push_pool_stats(&mut lines);
             push_server_stats(&mut lines);
             push_database_stats(&mut lines);
+            lines.push("".to_string()); // Ensure to end the stats with a line terminator as required by the specification.
 
             Response::builder()
                 .header("content-type", "text/plain; version=0.0.4")

src/query_router.rs

@@ -386,6 +386,18 @@ impl QueryRouter {
         }
     }
 
+    /// Determines if a query is a mutation or not.
+    fn is_mutation_query(q: &sqlparser::ast::Query) -> bool {
+        use sqlparser::ast::*;
+
+        match q.body.as_ref() {
+            SetExpr::Insert(_) => true,
+            SetExpr::Update(_) => true,
+            SetExpr::Query(q) => Self::is_mutation_query(q),
+            _ => false,
+        }
+    }
+
     /// Try to infer which server to connect to based on the contents of the query.
     pub fn infer(&mut self, ast: &Vec<sqlparser::ast::Statement>) -> Result<(), Error> {
         if !self.pool_settings.query_parser_read_write_splitting {
@@ -428,8 +440,9 @@ impl QueryRouter {
                     };
 
                     let has_locks = !query.locks.is_empty();
+                    let has_mutation = Self::is_mutation_query(query);
 
-                    if has_locks {
+                    if has_locks || has_mutation {
                         self.active_role = Some(Role::Primary);
                     } else if !visited_write_statement {
                         // If we already visited a write statement, we should be going to the primary.
@@ -1048,6 +1061,11 @@ impl QueryRouter {
         self.active_shard
     }
 
+    /// Set active_role as the default_role specified in the pool.
+    pub fn set_default_role(&mut self) {
+        self.active_role = self.pool_settings.default_role;
+    }
+
     /// Get the current desired server role we should be talking to.
     pub fn role(&self) -> Option<Role> {
         self.active_role
@@ -1113,6 +1131,26 @@ mod test {
         assert_eq!(qr.role(), None);
     }
 
+    #[test]
+    fn test_split_cte_queries() {
+        QueryRouter::setup();
+        let mut qr = QueryRouter::new();
+        qr.pool_settings.query_parser_read_write_splitting = true;
+        qr.pool_settings.query_parser_enabled = true;
+
+        let query = simple_query(
+            "WITH t AS (
+                SELECT id FROM users WHERE name ILIKE '%ja%'
+            )
+            UPDATE user_languages
+               SET settings = '{}'
+              FROM t WHERE t.id = user_id;",
+        );
+        let ast = qr.parse(&query).unwrap();
+        assert!(qr.infer(&ast).is_ok());
+        assert_eq!(qr.role(), Some(Role::Primary));
+    }
+
     #[test]
     fn test_infer_replica() {
         QueryRouter::setup();

tests/ruby/load_balancing_spec.rb

@@ -56,6 +56,41 @@ describe "Random Load Balancing" do
       end
     end
   end
+
+  context "when all replicas are down " do
+    let(:processes) { Helpers::Pgcat.single_shard_setup("sharded_db", 5, "transaction", "random", "debug", {"default_role" => "replica"}) }
+
+    it "unbans them automatically to prevent false positives in health checks that could make all replicas unavailable" do
+      conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+      failed_count = 0
+      number_of_replicas = processes[:replicas].length
+
+      # Take down all replicas
+      processes[:replicas].each(&:take_down)
+
+      (number_of_replicas + 1).times do |n|
+        conn.async_exec("SELECT 1 + 2")
+      rescue
+        conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+        failed_count += 1
+      end
+
+      expect(failed_count).to eq(number_of_replicas + 1)
+      failed_count = 0
+
+      # Ban_time is configured to 60 so this reset will only work
+      # if the replicas are unbanned automatically
+      processes[:replicas].each(&:reset)
+
+      number_of_replicas.times do
+        conn.async_exec("SELECT 1 + 2")
+      rescue
+        conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+        failed_count += 1
+      end
+      expect(failed_count).to eq(0)
+    end
+  end
 end
 
 describe "Least Outstanding Queries Load Balancing" do
@@ -161,4 +196,3 @@ describe "Least Outstanding Queries Load Balancing" do
     end
   end
 end
-