Continuous Integration

For the last few years there has been a slow growing improvement to the testing and QA Squid is subject to. This last week has seen the construction and rollout  of a full-scale build farm to replace some of our simple internal testing. Robert Collins covers the growth process in his blog.

Here is the initial release notice:

Hi, a few of us dev’s have been working on getting a build-test environment up and running. We’re still doing fine tuning on it but the basic facility is working.

We’d love it if users of squid, both individuals and corporates, would consider contributing a test machine to the buildfarm.

The build farm is at http://build.squid-cache.org/ with docs about it at http://wiki.squid-cache.org/BuildFarm.

What we’d like is to have enough machines that are available to run test builds, that we can avoid having last-minute scrambles to fix things at releases.

If you have some spare bandwidth and CPU cycles you can easily volunteer.

We don’t need test slaves to be on all the time – if they aren’t on they won’t run tests, but they will when the come on. We’d prefer machines that are always on over some-times on.

We only do test builds on volunteer machines after a ‘master’ job has passed on the main server. This avoids using resources up when something is clearly busted in the main source code.

Each version of squid we test takes about 150MB on disk when idle, and when a test is going on up to twice that (because of the build test scripts).

We currently test:

• 2.HEAD
• 3.0
• 3.1
• 3.HEAD

I suspect we’ll add 2.7 to that list. So I guess we’ll use abut 750MB of disk if a given slave is testing all those versions.

Hudson, our build test software, can balance out the machines though – if we have two identical platforms they will each get some of the builds to test.

So, if your favorite operating system is not currently represented in the build farm, please let us know – drop a mail here or to noc @ squid-cache.org – we’ll be delighted to hear from you, and it will help ensure that squid is building well on your OS!

-Rob

That just about covers everything. Hardware and build software requirements are listed in the build farm page.

Hi, a few of us dev's have been working on getting a build-test
environment up and running. We're still doing fine tuning on it but the
basic facility is working.

We'd love it if users of squid, both individuals and corporates, would
consider contributing a test machine to the buildfarm.

The build farm is at http://build.squid-cache.org/ with docs about it at
http://wiki.squid-cache.org/BuildFarm.

What we'd like is to have enough machines that are available to run test
builds, that we can avoid having last-minute scrambles to fix things at
releases.

If you have some spare bandwidth and CPU cycles you can easily
volunteer. 

We don't need test slaves to be on all the time - if they aren't on they
won't run tests, but they will when the come on. We'd prefer machines
that are always on over some-times on.

We only do test builds on volunteer machines after a 'master' job has
passed on the main server. This avoids using resources up when something
is clearly busted in the main source code.

Each version of squid we test takes about 150MB on disk when idle, and
when a test is going on up to twice that (because of the build test
scripts).

We currently test
2.HEAD
3.0
3.1
3.HEAD

and I suspect we'll add 2.7 to that list. So I guess we'll use abut
750MB of disk if a given slave is testing all those versions.

Hudson, our build test software, can balance out the machines though -
if we have two identical platforms they will each get some of the builds
to test.

So, if your favorite operating system is not currently represented in
the build farm, please let us know - drop a mail here or to noc @
squid-cache.org - we'll be delighted to hear from you, and it will help
ensure that squid is building well on your OS!

-Rob

Life of a Beta

From early inception when the developers have nothing but dreams for it.  Through the coding and arguments about what should be included and how. Through the alpha testing with its harrowing hours pondering obscure code from last decade. Even the odd period of panic as security bugs are whispered about behind closed doors. Such is the early life of software.

Two weeks ago word went out that 3.1 was reaching end-game.

This part of the release lifecycle seems to be going well. Packages appearing very slowly as QA throws demanding eyes on the code and making us actually fix things. Don’t be fooled by the packages out already, they have been in QA for a few months to get this far. On that note:

NetBSD, Gentoo, Ubuntu, FreeBSD and RedHat already have packages ready and available for at least testing use if you know where to look (ie the links right there might be a good start).

Debian has a bit more QA to go as of the writing, but the maintainer tells me there will be packages out soon.

OpenBSD and Mac turned out at the last minute to be running split-stack IPv6 implementations (for security apparently). All the documentation read in two years left the impression it was a Windows XP anarchism (and who runs XP Pro on a server?), so support was delayed and delayed.  The OpenBSD maintainer and someone interested from Mac are working with myself on closing that gap in the features.

There may be more OS with 3.1 packages. I’ve only begun working my way down the distrowatch.org popularity list to see which OS do and who to contact. Squid has bundles on over 600 OS apparently.

If you know who does the official packaging for your OS and whether there are 3.1 packages ready, please do me a favor and mention it. I’m seeking a web page where to find the squid (or squid3/squid30/squid31) package information and also the place where distro bug reports about Squid might end up.

Release 3.1

Kinkie pointed out Linus Torvalds blog today to the rest of us here working on Squid. As the release maintainer for Squid-3 this year I kind of agree, its a sad time to cutting a new version. For me its more of a reflection that for all the high hopes we have of this new release, we had the same or similar hopes of the earlier one. Just 12 months ago now.

On that sad note, yes its finally happened. 3.0 has aged into a full blown stable package. Most of a month and no new bugs. Perfect time for something shiny and new for the neo-tech fanclub. And so with that for an intro we are gone for 3.1 !

3.1 is available for beta testing in the form of 3.1.0.1. see the Release Notes for further details on the finer details of change.

This release has gained from the experiences of 3.0 and 2.6, starting from a much more stable base of code than the initial. 3.0 had a long period of years with few active developers, an interminably long period of testing releases, and in hindsight a premature birth.

Alongside the code this release has a wider collaboration with active users. For the first time in many years we held a Developer meeting that included Users. We who were there certainly took in a lot of feedback from all sides. I hope those users who talked to us can see in this release that their comments, even those made in passing, have been listened to and worked on.

The small comment from one user when asked what their biggest itch with squid was “we don’t like these being called STABLE, when its obvious they are not.” has led to the most notable change made to 3.1.  That comment and similar feelings by others lead us into discussions on the release naming and numbering. From which we have produced – 3.1.0.1 – the second milestone point of the branch we are calling 3.1. Where the developers have everything done and working for us.

no more DEVEL, PRE, or RC, no more premature labels guessing when things might be STABLE.  Just 3.1.0.1. Further testing from the rest of you will show whether anyone can consider it stable, unstable, usable or as buggy as raw earth.

From the developers; We use it. We love it. Try it, and see for yourselves.

Some of the stuff you will find there is;

• a lot of small changes aimed towards easier use and configuration (three cheers to those who nagged long an hard for this).
• a lot of network RFC compliance extensions, making 3.1 much more capable of meeting modern network needs. The future still holds improvements, but 3.1 is definitely better in many respects than everything that came before.
• a lot of things to make Squid a better experience for your own users. More seamless network recovery tricks than ever before. We have even tagged along behind the international localization bandwagon in our own way to make the errors squid does have to show both pretty and readable.

Sadly, careful readers will notice a section of the Release Notes labeled “Regressions against 2.7″.  Yes, those of you who moved to 2.7 because you needed some brand new feature there may still have trouble migrating up to 3.1. What we have done is to port as many of the 2.6 features and fixes as we could. A few did not make it in time, but will be coming in 3.2, alongside the features added as experimental in 2.7.

On the overview:

• 2.5 has disappeared over the horizon into the long dark night of obsoletion.
• 2.6 is itself officially aging out now. Supported, but the developer first response is “can you try something newer?”.
• 2.7 is being maintained for the few extremely high-performance accelerator setups. But in general the Squid-2 sequence is aging out for us developers.
• 3.0 has reached a point of stability, though not fully-featured.
• 3.1 is available for testing as the next step up. You should be planning to migrate up to 3.1 or later release.

If there are any features holding you to Squid-2, or even an issues you find with testing Squid-3 speak up, we rely on your input to choose the most needed features for porting.

Thank you all, and enjoy your use of Squid 3.1

Chunked Decoding

We have been getting a growing number of reports and bugs from people using Squid 3.0 described as ’squid producing a blank page’ when bypassing squid apparently works.

Sounds familiar to some yes? I’m bringing it up now because while it is an old problem, its not the TCP issues Adrian wrote about earlier and you should also check if you find its not this. Which incidentally can have exactly the same visible effects for end-users.

This ‘new’ issue is caused by certain widely-used web servers which shall remain nameless and unadvertised by me. Which always respond with HTTP/1.1 chunked-encoding of pages.

Servers are explicitly forbidden from sending that particular encoding type to software announcing itself as HTTP/1.0 (such as squid). But the broken server is doing it anyway!

Ironically: The authors use this server on their own help and support website. So those who are having this problem both see it as a squid problem, and can’t find or see any solution they may have posted anyway.

How to tell if this is your problem?

Use squidclient to make a web request that bypasses the squid proxy. It should send out the HTTP/1.0 request and get a page back. If the headers of the response include “Transfer-Encoding: chunked” there is your problem.

This is currently only an issue in Squid 2.5 or earlier and 3.0, which is still highly modeled around 2.5.

The solutions are varied depending on your capabilities.

Simplest for some will be to just bypass squid for those domains.

[ UPDATE: (thanks Michael Graham)

Apparently several people are having success with simply dropping the Accept-Encoding header to certain of these broken servers. Adding this to their squid.conf :

# Fix broken sites by removing Accept-Encoding header
acl broken dstdomain ...
request_header_access Accept-Encoding deny broken

NP: don't forget to remove it again when you upgrade out of 3.0

]

Next best is to use peer-routing to divert those domain requests at a squid 2.6 (or if you are feeling experimental a 3.1 build)

If its a serious issue and you are accelerating for one of these broken web servers. Then you will need to stick with Squid 2.6 until 3.1 is available for production use.

Why does it work for 2.6 and 3.1 but not 3.0?

Well, things are a bit messy I’ll have to write it up one day. Suffice to say that 3.1 has a lot more HTTP/1.1 support where the chunked-encoding/decoding was intended for. But 2.6 needed it a bit earlier so a version of the decoding (only!) was done to fit 2.6 needs at solving this same issue for high-performance users earlier last year.

The 3.0 code is just different enough that it would need a whole new back-port project to get it going well. The time and work that would take is being used instead to get 3.1 out faster. Which should be within a month of this writing so procrastinating could solve the problem for you.

[UPDATE: Thanks to the Gentoo Project for their work back-porting this will be available from 3.0.STABLE16-RC1 ]

Squid-2 performance work: graph #1

 

Whats going on with Squid-2 and Squid-3 ?

A few people have asked me what the deal is with Squid-2 and Squid-3.

“Why are you developing on Squid-2 when Squid-3 is now out?”

“Should I upgrade to Squid-3 now that its released?”

I’m focusing on Squid-2 for a few reasons, namely:

• Its what people running high-traffic sites are currently running, and Squid-3 doesn’t work at all for them;
• I was fed up waiting for Squid-3 to be released and for it to become mature enough for users to migrate to before I started my performance work. I gave up about 12 months ago and began planning out the work thats currently going on.
• I’m personally much more familiar with the Squid-2 codebase than the Squid-3 codebase.

So what exactly am I doing to Squid-2? Well, I’m doing all the things to Squid-2 which I personally believe we should’ve done in the C++ Squid-3 branch before all the “new stuff” was added. You can find it all at http://devel.squid-cache.org/changesets/squid/s27_adri.html . A summary of what I’m doing in this first round:

• I’m taking a very sharp scalpel to the codebase and removing all of the extra data copies and buffering which is going on;
• I’m reworking the buffer management so arbitrary sized data buffers can be used, rather than fixed 4k buffers for network/disk traffic;
• I’m reworking the Strings interface to use reference counting and reference underlying buffers, saving on memcpy() and malloc() calls, cutting down on the amount of transient memory used to handle requests and dropping the CPU and memory bus utilisation quite dramatically;
• I’m reworking the dataflow between server->store and store->client to use the above reference counted buffers, so data isn’t memcpy()’ed between layers, again dropping CPU and memory bus utilisation;
• And I’m going to break out as much of the code into external libraries with well-understood dependencies, as preparation for documentation, unit testing and further profiling.

My aim is to fix whatever bugs show up in Squid-2.7 and then in Squid-2.HEAD (which has some of the above included already.) I’ll then start bringing across my changes as they’ve been tested and been found stable. My aim is to have the bulk of the above done within the next month or so and get it into Squid-2.HEAD and concentrate on making it stable before I continue tidying up the dataflow and restructuring the ugly bits of code.

Whats this mean for Squid-3? The Squid-3 guys are doing some great work with things such as ICAP and IPv6 and I hope that they’ll gain more experience with their codebase over the next 12 months or so. I’m certainly not bringing ICAP support into Squid-2 until I’ve reworked the dataflow and tidied up the code enough for ICAP to sit comfortably in the data pipeline, rather than have it bolted onto the side and hooking into strange places where it shouldn’t. (I may bring in IPv6 into Squid-2 soon though!)

Hopefully my work and their work will culminate with the development of the next Squid major version over the next 12 to 24 months. There’s a long way to go though and my main aim here is to get faster, better and shinier code out to the majority of Squid users now so they can benefit from the development, rather than repeating the 4-odd year gap between Squid-2.5 and Squid-2.6. Users hated that.

So whats it mean for you?

• If you want to try out Squid-3; if you want supported ICAP services then try it out.
• Squid-2.X will continue being developed over the next 12 months as time permits, so don’t feel like you have to move to Squid-3.
• If you feel adventurous, try out Squid-2.7. Initial reports are that its stable and slightly less CPU intensive.
• Squid-2.7 is the first version to include changes to allow Youtube and Microsoft Updates caching. It doesn’t do it out of the box, but the support is there, and I’ll be publishing test rules soon to let people start caching this stuff.
• If you feel really adventurous then try out Squid-2.HEAD and report back if you have any issues. It should be even less CPU intensive, but only under certain workloads.

How cachable is google (part 2) – Youtube content

Youtube is (one of) the bane of small-upstream network administrators. The flash files are megabytes in size, and a popular video can be downloaded by half the people in the office or student residential college in one afternoon.

It is, at the present time, very difficult to cache. Lets see why.

There’s actually two different methods employed to serve the actual flash media files that I’ve seen. The first method involves fetching from youtube.com servers; the second involves fetching from IP addresses in Google IP space.

The first method is very simple: the URL form is:

http://XXX-YYY.XXX.youtube.com/get_video?video_id=VIDEO_ID

XXX is the pop name; YYY is I’m guessing either a server or a cluster name.

This is pretty standard stuff – and If-Modified-Since requests seem to also be handled badly too! The query string “?” in the URL makes it uncachable to Squid by default, even though its a flash video. Its probably not going to change very often.

The second method involves a bit more work. First the video is requested from a google server. This server then issues a HTTP 302 reply pointing the content at a changing IP address. This request looks somewhat like this:

http://74.125.15.83/get_video?video_id=HrLFb47QHi0&origin=dal-v37.dal.youtube.com

Again, the “?” query string. Again, the origin, but its encoded in the URL. Finally, not only are If-Modified-Since requests not handled correctly, the replies include ETags and requests with an If-None-Match revalidation still return the whole object! Aiee!

So how to cache it?

Firstly, you have to try and cache replies with a “?” reply. It would be nice if they handled If-Modified-Since and If-None-Match requests correctly when the object hasn’t been modified – revalidation is cheap and its basically free bandwidth. They could set the revalidation to be, say, after even 30 minutes – they’re already handling all the full requests for all the content, so the request rate would stay the same but the bandwidth requirements should drop.

The URLs also have to rewritten, much like they do to cache google maps content. The “canonical” form URL will then reference a “video” regardless of which server the client is asking.

Now, how do you do this in Squid? I’ve got some beta code to do this and its in the Squid-2 development tree. Take a look here for some background information. It works around the multiple-URL-referencing-same-file problem but it won’t unfortunately work around their broken HTTP/1.1 validation code. If they fixed that then Youtube may become something which network administrators stop asking to filter.

(ObNote: the second method uses lighttpd as the serving software; and it replies with a HTTP/1.1 reply regardless of whether the request was HTTP/1.0 or HTTP/1.1. Grr!)

How cachable is google (part 1): Google Maps

I’m looking at how cachable Google content is with an eye to make Squid cache some of it better. Contrary to popular belief, a lot of the google content (that I’ve seen!) is dynamically generated “static” content – images, videos – which could be cached but unfortunately aren’t.

Google Maps works by breaking up the “map” into multiple square tiled images. The various compositing that occurs (eg maps on top of a satellite image) are rendered by the browser and not dynamically generated by Google.

We’ll take one image URL as an example:

http://kh3.google.com.au/kh?n=404&v=23&t=trtqtt

A few things to notice:

1. The first part of the hostname – kh3 – can and does change (I’ve see kh0 -> kh3.) All the tiles as far as I can tell can be fetched from each of these servers. This is done to increase concurrency in the browser: the Javascript will select one of four servers for each tile so the concurrency limit is reached for multiple servers (ie, N times the concurrency limit) rather than just to one server.
2. The query string is a 1:1 mapping between query and tile, regardless of which keyhole server they’re coming from.
3. The use of a query string negates all possible caching, even though…
4. .. the CGI returns Expires and Last-Modified headers!

Now, the reply headers (via a local Squid):

HTTP/1.0 200 OK
Content-Type: image/jpeg
Expires: Sat, 15 Nov 2008 02:44:29 GMT
Last-Modified: Fri, 17 Dec 2004 04:58:08 GMT
Server: Keyhole Server 2.4
Content-Length: 15040
Date: Fri, 16 Nov 2007 02:44:29 GMT
Age: 531
X-Cache: HIT from violet.local
Via: 1.0 violet.local:3128 (squid/2.HEAD-CVS)
Proxy-Connection: close

The server returns a Last-Modified header and Expires header; but as it has a query identifier in the URL (ie, the “?”) then plenty of caches and I’m guessing some browsers will not cache the response, regardless of the actual cachability of the content. See RFC2068 13.9 and RFC2616 13.9. Its unfortunate, but what we have to deal with.

Finally, assuming the content is cached, it will need to be periodically revalidated via an If-Modified-Since request. Unfortunately the keyhole server doesn’t handle IMSes correctly, always returning a 200 OK with the entire object body. This means that revalidation will always fail and the entire object will be fetched in the reply.

So how to fix it?

Well, by default (and for historical reasons!) Squid will not cache anything with “cgi-bin” or “?” in the path. Thats for a couple of reasons – firstly, replies from HTTP/1.0 servers with no expiry information shouldn’t be cached if it may be a CGI (and “?”’s generally are); and secondly intermediate proxies in the path may “hide” the version of the origin server and you never quite know whether it was HTTP/1.0 or not.

Secondly, since the same content can come from one of four servers:

• You’ve got a 1 in 4 chance that you’ll get the same google host for the given tile; and
• You’ll end up caching the same tile data four times.

I’m working on Squid to work around these shortcomings. Ideally Google could fix the second one by not using query-strings but instead using URL paths with correct cachability information and handling IMS, eg:

http://kh3.google.com.au/kh?n=404&v=23&t=trtqtt

might become:

http://kh3.google.com.au/kh/n=404/v=23/t=trtqtt

That response would be cachable (assuming that they didn’t vary the order of the query parameters!) and browsers/caches would be able to handle that without modification.

I’ve got a refresh pattern to cache that content but its still a work in progress. Here’s an example:

refresh_pattern    ^ftp:            1440 20% 10080
refresh_pattern    ^gopher:    1440 0% 1440
refresh_pattern    cgi-bin        0 0% 0
refresh_pattern    \?                0 0% 4320
refresh_pattern    .                    0 20% 4320

I then remove the “cache deny QUERY” line and simply use a cache allow all; then I use refresh_pattern’s to match on which patterns shouldn’t be cachable if no expiry information is given (ie – if a URL with cgi-bin or ? in the path returns expiry information then Squid will cache it.)

[UPDATE: We have now merged the results of Adrians work here into Squid-2.7 and 3.1+. The new requirement for refresh_patterns are:

refresh_pattern    ^ftp:        1440  20% 10080
refresh_pattern    ^gopher:        1440   0% 1440
refresh_pattern    -i (/cgi-bin/|\?)        0   0% 0
refresh_pattern    .        0   20% 4320

hierarchy_stoplist cgi-bin ?

]

It’d then be nice if Google handled IMS requests by the keyhole server correctly!

Secondly, Squid needs to be taught that certain URLs are “equivalent” for the purposes of cache storage and retrieval. I’m working on a patch which will take a URL like this:

http://kh3.google.com.au/kh?n=404&v=23&t=trtqtt

Match on the URL via a regular expression, eg:

m/^http:\/\/kh(.*?)\.google\.com(.*?)\/(.*?)$/

And mapping that to a fixed URL regardless of the keyhole server number, eg:

http://keyhole-server.google.com.au.SQUIDINTERNAL/kh?n=404&v=23&t=trtqtt

The idea, of course, is that there won’t ever be a valid URL normally fetched whose host part ends in .SQUIDINTENRAL and thus we can use it as an “internal identifier” for local storage lookups.

This way we can then request the tile from any kh server ending in any country, so the following URLs would be equivalent from the point of view of caching:

http://kh3.google.com.au/kh?n=404&v=23&t=trtqtt
http://kh2.google.com.au/kh?n=404&v=23&t=trtqtt
http://kh0.google.com/kh?n=404&v=23&t=trtqtt

Its important to note here that the content is still fetched from the requested host, its just stored in the cache under a different URL.

I’ll next talk about caching Google Images and finally how to cache Youtube.

Why even bother making cachable content?

I see so many sites pop up in some Squid logs which seem to try and avoid any attempt at caching. I’m not sure why, but I’m going to try and cover a few points here.

1. I want to know exactly how many bits I’m shipping! This is especially prevalent in the American internet scene. Everyone’s about shipping bits. The more bits you ship the “better” you are. (There’s some talk about the “number of prefixes you advertise” also being linked to how “big” your network is; or maybe people are just lazy at trying to aggregate their BGP announcements. I digress..) Sure, if you graph your outbound links this is true. But you can do HTTP tricks to know exactly how many requests you’re handling without shifting the whole object out. Just set the objects to “must revalidate” rather than being immediately expired; let the web cache always revalidate the request via an If-Modified-Since request. You’ll get the IMS and can send back a “not-modified” reply; you can then synthesise a graph based on what you -would- be serving. Voila, free bits. This can be quite substantial if you have lots and lots of images on your site.
2. I want to know how many people are accessing my site! This is definitely a left-over from the 90s and even then the problem was solved. If you absolutely positively need to know about page impressions then just embed a non-cachable 1×1 transparent gif somewhere where it won’t slow the page rendering down. Leave the rest of the site cachable. Really though, these days people should just use javascript and cookies (a la the Google “urchin”) if they want accurate “people” and “impression” counts. Trying to do it based on page accesses and unique IPs just isn’t going to cut it.
3. I don’t want people to cache the data; they have to login first! You can tell proxy caches that they must first revalidate the authentication information from the origin server before serving out content. You can have your cake and eat it too.
4. Making my content cachable is too damned hard! How do I know what headers when and where? Its not all that difficult. Mark Nottingham’s Caching Tutorial covers a lot of useful information about building cachable websites. You can keep control of your authenticated content and push out more content than you’re actually buying transit for.

Just remember a few simple rules:

• Don’t hide static content behind query URLs (ie, stuff with a ‘?’ in them). Caches won’t cache them (unless, of course, they’re built by me. But then, I am pretty evil.) I see plenty of websites which hide all of their images and flash videos behind a CGI script with a ? in the path – caches just won’t bother trying to cache it. Amusingly, most of those sites hide static content behind CGI scripts! Just imagine what it’d be like to be able to push five or ten times the amount of content to clients behind proxy caches.
• Don’t be afraid to ask for help in how to optimise your site for forward caching. Heck, even asking on the squid-users mailing list will probably get you sorted out without too much trouble.
• There are people behind proxy caches – the developing world for one, but there’s plenty of caches to be found in schools, offices, wired buildings, wireless mesh networks and the like. Bandwidth isn’t free and never will be. You might be able to buy a 40gbit pipe to your favourite transit provider in North America but that won’t help people in South Africa or Australia where international bandwidth is still expensive and will remain so for the forseeable future. And yes, we like watching Youtube as much as the next person.

Squid-2.6.STABLE16 is out!

Henrik has released Squid-2.6.STABLE16. This resolves a number of bugs, including a crash bug introduced in Squid-2.6.STABLE15.

The changeset list explains whats changed; the release page includes downloads and other useful stuff. Don’t forget to read the release notes if you’re updating from 2.5 to 2.6!

And don’t forget the Squid-2.6 Configuration Manager!