- From: Willy Tarreau <w@1wt.eu>
- Date: Thu, 26 Jan 2012 15:01:47 +0100
- To: Adrien de Croy <adrien@qbik.com>
- Cc: Poul-Henning Kamp <phk@phk.freebsd.dk>, Amos Jeffries <squid3@treenet.co.nz>, ietf-http-wg@w3.org
On Fri, Jan 27, 2012 at 02:04:51AM +1300, Adrien de Croy wrote: > In our case, we always send complete chunks. At the point where we > decide to send a chunk, we have some data, and we know its length, and > we have decided to send it. > > So the signal it would always fall on the start of a new chunk, e.g. we > scan, determine the content to be malware, and so the next chunk sent > would be 0 and signal an abort due to malware. OK, I see, this is because you're doing deep content analysis. > A close would then be less ambiguous, since nothing explicit was > communicated, the client could then more reliably assume it was a > transient network error. > > I can't really think of a requirement to abort part way through a > chunk. Not in this scenario anyway. Probably in your case you never emit truncated chunks, but over the wire it happens for a variety of reasons, including timeouts and TCP aborts via active components (TCP proxies and/or firewalls) as I indicated in a previous example. > If the intermediary is re-chunking then it can easily get around this issue. But many have no reason to re-chunk if they're just forwarding data. Willy
Received on Thursday, 26 January 2012 14:02:35 UTC