Cloud Native Computing Foundation EnvoyCon NA 2022, 28 Oct 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Lightning Talk: Manipulating HTTP Headers Using a Full Set of Substitution... - Christoph Pakulski

Description

Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon Europe 2023 in Amsterdam, The Netherlands from April 17-21. Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.

Lightning Talk: Manipulating HTTP Headers Using a Full Set of Substitution Formatters - Christoph Pakulski, Tetrate

Envoy allows for modifying HTTP headers when sending requests upstream and responses downstream. The syntax used for creating header content is the same as for creating access log’s entry. But surprisingly, only a small subset of formatters were available for header modification compared to the full set of access log’s formatters. This has changed and now all formatters can be used in both, in header manipulation and in access logs. This talk describes logic used when creating access logs and when manipulating headers. It also explains why using the same formatter in the access log can render different results than using it in the header. Note: This feature is still under development, but I hope that it will be finished and merged to main before the conference. See https://github.com/envoyproxy/envoy/pull/21932

A

Trade and today I'll be talking about manipulating HTTP headers using full set of substitution formatters, but before we start talking about start talking about manipulating HCB headers, let's spend some time talking about access logs.

A

So when Android processes, transactions I mean it receives, a requests from down. Sim client selects Upstream host transmits the request and waits for the response and eventually sends it back to the client. It produces a lot of data and the data is extremely valuable. When things go wrong, you would like to know what happened, how many times envoir transmitted requests Upstream?

A

How was the time it was waiting for the response, so the easiest way to extend the data is the Via access logs now, access log is basically a string representation of some data which has been produced during those transactions and it's defined by a sequence of substitution, formatters and substitution. Formatter is a string which starts and ends with percentage signs and uniquely identifies the data you would like to display. So here I give an example of couple of uh of substitution formatters the start time and transaction basically started.

A

The second one is next: I would like to display a certain header from the request and the Response Code, which was received from from an upstream.

A

So now, let's move to manipulating hdb headers. So there's a mechanism which has been in Android for a while and basically what it allows you to do is to modify the headers as they pass invoice on the Upstream, Direction and downstream Direction, and so you can manipulate request headers on the response. Headers and the way to do it is via configuration. You have to specify two things. One is the name to give to the header to be added, and also you would like to identify the value so for identifying the value. As you see here.

A

Is uh we use again the substitution formatter as a string with stars and ends with a percentage sign and identifies the value which would like to add as a header to that header, um it looks exactly like access log, substitution, formatter. So the question is natural question is: is it the same thing? Are we talking about exactly the same thing so until now, the answer is no, even though the syntax is exactly the same as the year for Access log and for header manipulation, I can use the substitution formatting for hostname, but it has been.

A

It's process is passed and it's actually converted to a string using completely to different code. uh That, obviously, is not super optimal. We did. There is a code duplication, yeah to document it into different places. Testing efforts doubles. So that's from developer point of view from user perspective. Things are not much clearer, because uh when you look for a specific keyword, it will be directed into different places, one for Access log, formatters and one for header manipulation and to make things even more confusing for Access logs.

A

There is a set of about 80, plus formatters and for hidden manipulation is only about 30 which are available, and so there were questions like uh you know. Why can't why I cannot use a specific uh substitution formatter for manipulating how the why the same formatter is available for uh access logs.

A

So the answer is because the code was different. There was a different person and there were different implementation of the of the formatter itself, so starting with release, 124 Things become much more Unified, so the same parser is used both for Access long, the same for either manipulation and the same implementation of the substitution formatter and the substitution format that I mean by substitution. Formatter I mean uh the code which takes data and converts it to the string.

A

There will be some little bit differences as you see on the for Access log and header manipulation for the code and for a test, but they mostly uh limited towards processing the configuration, because access log is defined in a different way than header manipulation.

A

That code, it must be has to be different and testing. Routines are also different for the user. Things become much more easier too, because uh it all those uh substitution formatters are defined only in my documented only in one place and whenever you look for a specific keyword, they will be only one result.

A

So you may say: Hey That's great, so whenever I can use the access log, formatted I can also use it to manipulate the headers. But the answer is no and the difference between why one can be used in Access log and one cannot be used into the header. Manipulator is in timing.

A

When data is created so for Access, logs access logs are created when transaction finishes so, and we received the request from the client selected Upstream, send the request, waited for response and then transmitted to the downstream and then access log is created and, as you see at that moment, other data or data related to the transactions is already available because the event which happened in the past- but that's not true- for manipulating the headers.

A

So, for example, here I would like to manipulate the request headers right, but the only data which is available to display at this moment what to use is related to clients, connection, Downstream, client connection to Envoy and actually sending the request itself.

A

All the other things like you know, selecting Upstream, sending waiting for responses. Those are future events. They has not happened yet so using that uh using formatters, which relate to those events doesn't really make sense, and here is an example right and we try to manipulate the request headers, but the formatter relates to event in the future.

A

I would like to get some data from response, which obviously doesn't happen yet so in those cases you can still put them, but the result will the results produced by a formatter will be empty string, so things are a little bit better. You know when modifying response headers, mostly because the more data is available. Data related to Upstream connection, which host has been selected, uh whether the connection was TLS. How many times then we tried to reach Upstream um the time waited for response.

A

All those things are available right now, so here I give an example of Upstream request. Attempt I would like to put into a header how many times and if I tried to reach Upstream, uh but certain data is still not available. The data which is related to a future event, I mean transmitting uh response to the client.

A

So here is an example of the byte send how many bytes are sent back and we send back to um to the client. This is his future event and it does not produce uh the value which, which we expect so as a final thought, I would like to present the use case, which was not possible to do in uh prior to release 124.

A

So there is a network of a mesh network of of envoys and they select text Hub based on the random algorithm and I, would like this user to know which one has been taken.

A

So in order to do it, each Enterprise has to be configured with manipulating response headers by adding his own name on top of whatever was received from Upstream and as a response program, you know progresses from option to Downstream. Each Envoy adds its own name so here when the client eventually gets it it's easily to identify the path every time I send, it might be a different path. Super easy to debug all right. That's all! Thank you very much.