For both apps and configuration

As far as I understand pgagroal implements session level pooling, i.e. number of sessions is equal to number of Postgres backends.
So pgagroal can not be used to restrict number of Postgres backends (as pgbouncer/Odyssey can do). It can only help when clients do not establish permanent connections with database but perform connect/disconnect on each operation. But it seems to be quite exotic use case now, doesn't it?

Are there any plans to support transaction level pooling?

Add support for a daemon process, pid_file configuration parameter and a pgagroal-cli stop command

If hba_type is SECURITY_ALL we need to override it to auth_type in the pooled connection code path

Requires updates to SASL prep

Describe the architecture in more detail with all major components

• Shared memory
• Atomic operations
• Management
• libev usage

pgagroal-cli status should return

• Status of pgagroal
• Active connections
• Total connections
• Max connections

First round should be

• Network mask
• User name
• Database

Send a terminate message (X) upon connection close, if possible.

Set the process title for all connections such that ps will list them nicely.

Should only be

... pgagroal: myuser <active>

or

... pgagroal: myuser <in-pool>

such that the overhead of setting the status is limited.

Log connections and disconnections using log_connections and log_disconnections.

Both are Bool and off by default.

Like pgagroal-cli status but includes information of the potential "sub-pools" as well.

Support TLS v1.2 and TLS v1.3 transport

Add a cancel-shutdown command to get out of the GRACEFULLY state.

Create a lifecycle for a pipeline that allows to have a pipeline specific shared memory segment across instances

Move prefill() such that user connections can happen at the same time.

Support graceful shutdown where existing connections are allowed to finish their session.

Add gracefully command to pgagroal-cli and trigger on SIGTRAP as well.

Implement a SCRAM-SHA-256 security module in order to support prefill and pooling.

Should be based on a plugin architecture. Start with

• Script based (.sh files)

We need to send a bad password message in case we receive bad credentials for a pooled connection

Both for main and child processes

Add functionality to help create .rpm archives

The encrypted password string may contain characters that strlen() doesn't understand.

Pass the string length in explicit.

Add an option to the configuration to explicit set the desired mode of libev

Add a -U switch for specifying a user name, and a -P switch for specifying a password.

These switches should be used in scripted environments.

Implement log rotation for files - could be based on a date or size.

Will require log_path to understand date modifiers like %Y, %m, %d, %H, %M and %S.

Create a vault where user names and passwords are stored

Create a pipeline that pools connections on the transaction boundary.

The slot will be valid during the start() method of the pipeline. The user / database information should be copied to a pipeline memory segment.

If pgagroal_get_connection() returns a non-zero value an error should be sent to the client and communication terminated.

Implementation should be in pipeline_transaction.c.

Configuration should be done through

[pgagroal]

pipeline = transaction

To maintain code style across the project

http://uncrustify.sourceforge.net/

Implement log formatting codes that can style the logging output. Could borrow similar codes as PostgreSQL.

May require changes to zf_log.

This should be done as a separate pipeline

Add a pgagroal_databases.conf file that specifies connection limits for a database, a user or both.

Add an option for connection validation

• off
• foreground
• background with interval