Re: Health checks: check for string in return data?

From: Nate Carlson <haproxy#natecarlson.com>
Date: Mon, 30 Jun 2008 15:00:33 -0500 (CDT)


On Mon, 30 Jun 2008, Willy Tarreau wrote:
> I've also received this request. This needs a more general rework of the
> health-checks, because right now, only the beginning of the response is
> read, but we cannot say that there is a proper "parsing". Some people
> want to be able to consider a check as failed if the find a regex in the
> response, or if they don't find a regex in the response. But since there
> is no unlimited buffer, the check may either scan a very small chunk
> reliably, or scan a larger one read at once in a shared buffer.

Nods, makes sense.

> Do you know where in your response is the string you're looking for ? Is
> it close to the beginning or close to the end, and what is the size of
> the response ?

In my case, it's always a short response (under 512 bytes), and is at the beginning of the returned page. It's always hitting a custom-built page for this purpose.

> I'm asking, because maybe we could consider that for plain text, we may
> scan the whole response even in small chunks, if we always keep a buffer
> as large as the string we're looking for.

Yeah, I think that would make sense.



| nate carlson | natecars@natecarlson.com | http://www.natecarlson.com | | depriving some poor village of its idiot since 1981 |
Received on 2008/06/30 22:00

This archive was generated by hypermail 2.2.0 : 2008/06/30 22:16 CEST