not logged in | [Login]

The SQL module needs to be async. Which involves a lot of changes. Philippe has start on changes


The underlying xlat code is still synchronous. That needs to be updated before the SQL module will be full async.


As with XLats, the code is still synchronous. That needs to be updated before the SQL module will be full async.


The paircompare callbacks (e.g. SQL-Group) are still sync


The new V4 code doesn't have a MOD_SESSION section so the simultaneous session checking needs to be done differently.

We probably should have a dedicated virtual server for this. Needs to be written.

NAS from SQL

This is done during the instantiation phase of the module. We would need major changes to make it async.

After some discussion, it's probably best to remove the "read clients" code from SQL.

i.e. the "NAS from SQL" functionality should be replaced by the "dynamic clients" code. We now have the map functionality in the server, so NAS information can be read from the nas table. That should simplify the module.

The async nature of v4 means that the server should be able to accept a large number of packets from unknown clients, and run those through the "dynamic clients" code.

The requirement here is that the v4 network stack be able to track multiple outstanding packets from "unknown" clients. In v3, it is hard-coded to only allow one packet from an unknown client.

Query timeout

Currently, query timeouts are only supported for MySQL. The new async code makes it easier to support this for Postgres too but the question is: "what do we do when the timer is triggered?"

  • One option is to cancel the query, but the cancel API is synchronous which makes it impossible to use.
  • Another option is to ignore the query (let it finish) and continue processing the request but this means the socket remains unusable.
  • The last option is to simply close the socket and continue processing the request

The last option is probably the best as there's no point in letting the query finish.

IP pool modules

Both rlm_sqlippool & rlm_sqlhpwippool use some functions from the SQL module. They need updating to be compatible with the new async model.

Instead of sharing the connection pool with the sql module instance as is the case now, we could share the thread specific data from the sql module to access the sql module's thread pool. Otherwise, it means having our own pool.