[prev in list] [next in list] [prev in thread] [next in thread] 

List:       freeradius-users
Subject:    Re: [EXTERNAL] Performance issues
From:       Alan DeKok <aland () deployingradius ! com>
Date:       2022-10-05 1:26:47
Message-ID: B6702483-95A5-4AC4-95EF-5876FA020C12 () deployingradius ! com
[Download RAW message or body]

On Oct 4, 2022, at 9:45 AM, Winfield, Alister (Senior Solutions Architect) via \
Freeradius-Users <freeradius-users@lists.freeradius.org> wrote:
> 
> Remember that the query may take <1s but the time for the query response at the \
> client can be >5s. Its often because there is a queue built up of requests and the \
> DB is too slow to service them thus the ‘enqueued' ones are around for too long. \
> Fixing is mostly make the DB faster eg by adding another server and synchronising \
> the data, fixing indexes, ensuring queries aren't idiotic (eg full table scans). \
> Sometimes it's the network eg TLS using a dumb slow random number source that's not \
> getting enough entropy so blocking or perhaps might be packet loss, etc etc etc.

 What happens in most of these discussions is the advice to specifically check about \
5 different things.  DB size, indexes, etc.  When the response is "I only checked one \
thing", that's a recipe for continued problems.

> Sadly, this is all about analysing the system bit by bit until you find the real \
> issue and fixing that.. For your sanity start at the bottom of the stack and work \
> up. FreeRadius is highly unlikely to be the problem. It comfortably handles 30K \
> requests per second and without much effort can do much more.

 Yup/

 It's not that the "SQL slow query log" is lying.  It's not.  But it's also not \
testing end-to-end functionality.  There's a lot more going on in the network, SQL \
database, etc.

 The FreeRADIUS "SQL module is blocked" message *is* testing end-to-end \
functionality.  It's testing *all* of the time from "FreeRADIUS asked SQL a question" \
to "SQL returned an answer 10 seconds later".

 That log message *never* lies, and it *always* indicates a problem.

 Alan DeKok.
-
List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html


[prev in list] [next in list] [prev in thread] [next in thread]