Add support for multi-database / multi-user pools #96
Conversation
| return Ok(Client { | ||
| read: BufReader::new(read), | ||
| write: write, | ||
| addr, | ||
| buffer: BytesMut::with_capacity(8196), | ||
| cancel_mode: true, | ||
| transaction_mode: transaction_mode, |
There was a problem hiding this comment.
cancel mode don't care about transaction_mode so it is easier to hard code a value instead of working our way backwards from the secret to the corresponding pool
src/client.rs
Outdated
| target_pool = match get_pool(database.clone(), user.clone()) { | ||
| Some(pool) => pool, | ||
| None => { | ||
| error_response( | ||
| &mut write, | ||
| &format!("No pool configured for database: {:?}, user: {:?}", database, user), | ||
| ).await?; | ||
| return Err(Error::ClientError) | ||
| } | ||
| }; | ||
| transaction_mode = target_pool.settings.pool_mode == "transaction"; |
There was a problem hiding this comment.
This concept of target_pool simplifies the implementation. Once a client connects it is permanently assigned a target_pool. This target_pool connects to one cluster (with whatever shards/instances configured for that cluster) using a specific user. Almost everything after more or less behaves identically to how Pgcat operated before
| let correct_user = config.general.admin_username.as_str(); | ||
| let correct_password = config.general.admin_password.as_str(); |
There was a problem hiding this comment.
This is new. Instead of any authenticated user being allowed to access admin db, a specific username/password is required for that. Honestly, I went with that for simplicity
| @@ -410,7 +448,7 @@ where | |||
|
|
|||
| // The query router determines where the query is going to go, | |||
| // e.g. primary, replica, which shard. | |||
| let mut query_router = QueryRouter::new(); | |||
| let mut query_router = QueryRouter::new(self.target_pool.clone()); | |||
There was a problem hiding this comment.
Query routing configs are per-pool now and not global so the router needs access to the pool configs
src/pool.rs
Outdated
| pub static POOLS: Lazy<ArcSwap<PoolMap>> = | ||
| Lazy::new(|| ArcSwap::from_pointee(HashMap::default())); |
src/admin.rs
Outdated
| let value = stats.get(column.0).unwrap_or(&0).to_string(); | ||
| row.push(value); | ||
| for (_, pool) in get_all_pools() { | ||
| let pool_config = pool.settings.clone(); |
There was a problem hiding this comment.
I think you can borrow the pool config here, e.g. let pool_config = &pool.settings;, but it doesn't really matter.
src/client.rs
Outdated
| } | ||
|
|
||
| debug!("Password authentication successful"); | ||
|
|
||
| auth_ok(&mut write).await?; | ||
| write_all(&mut write, get_pool().server_info()).await?; | ||
| backend_key_data(&mut write, process_id, secret_key).await?; | ||
| if !admin { |
There was a problem hiding this comment.
You can send this for admin users too, it won't hurt I think.
There was a problem hiding this comment.
For admin, there isn't really any server, and the pool we use to get server_info is a dummy one.
There was a problem hiding this comment.
Yeah, but if you don't send server info, the psql will look weird. We prewarm the pool on boot, so we actually have server info, so we have something to send to the client.
src/config.rs
Outdated
|
|
||
| for (pool_name, pool_config) in &self.pools { | ||
| info!("--- Settings for pool {} ---", pool_name); | ||
| info!("Total Pool size from all users: {}", pool_config.users.iter().map(|(_, user_cfg)| user_cfg.pool_size ).sum::<u32>().to_string()); |
There was a problem hiding this comment.
| info!("Total Pool size from all users: {}", pool_config.users.iter().map(|(_, user_cfg)| user_cfg.pool_size ).sum::<u32>().to_string()); | |
| info!("Pool size from all users: {}", pool_config.users.iter().map(|(_, user_cfg)| user_cfg.pool_size ).sum::<u32>().to_string()); |
No description provided.