Starting up GitHub sponsors and some recent postings work

Hello everyone! I am happy to announce that I’ve set up GitHub sponsors on my profile. If you want to support my blog or my work on Thanos/Prometheus, and you have some free money then now you have a way to throw some money at these projects. Let’s see if I will even get one sponsor. I was thinking that maybe I should work on some custom features that could be behind a paywall. Let’s see when I will have some time to work on them.

I haven’t written anything on my blog for quite some time. I think it’s high time I’ve revived it. Writer’s paralysis probably happened to me, so I haven’t posted anything. Somehow I kept thinking about many topics but was afraid of writing about them and clicking “Publish”. But now it’s time to not be afraid and do that 🙂

Probably the most exciting stuff that I have worked on (and still do) recently is postings encoding improvements in Prometheus & Thanos. It’s now possible to specify a custom postings encoder in the Prometheus compactor: https://github.com/prometheus/prometheus/pull/13242. After https://github.com/prometheus/prometheus/pull/13567 it will even be possible to use a custom postings decoder. The postings data structure sits at the core of the Prometheus TSDB – it is used for storing sets of sorted integers. Whenever someone specifies some label matcher in a query e.g. {foo="bar"} then Prometheus goes through the set of series (postings) which have foo="bar" in their labels. So, it is paramount to make this data structure as efficient as possible.

Currently, each integer is simply stored using 4 bytes. It’s possible to be much better than that. For example, if you have a set of integers 1, 2, 3, 4..., 10 then it’s enough to only say that there’s a run of 10 integers starting from 1. Over time, many more techniques for compression were invented.

I have researched what is available and found out that the most popular paper (probably) is this one https://arxiv.org/abs/1401.6399 by Daniel Lemire & others. I love his work in particular because he always puts up the source code for his paper. It’s a huge help! I wish more people had done that.

We have a few constraints in the Thanos/Prometheus world:

  • We should read posting lists only in one direction i.e. we shouldn’t need to read them twice. Some encoding formats force the reader to read twice like the patch frame-of-reference variants. This constraint is needed because we would like to avoid allocating memory for the whole postings list if possible to save a lot of memory. In the Thanos world, the list could be easily hundreds of millions in size.
  • The intersection must be very fast. Prometheus/Thanos will do intersections many more times than encode/decode data. It’s not uncommon to have 3+ label matchers in a single query.

From all of the things I’ve looked at, S4-BP128-D4 and roaring bitmaps look the most promising. The latter is used by a lot of similar projects already like M3DB. The former might be not so popular but it is specifically designed for SIMD which gives us very fast encoding/decoding.

I even started writing a Go version of S4-BP128-D4 but I haven’t finished it, yet: https://github.com/GiedriusS/go-bp. So, I am opting to try roaring bitmaps first. Even then it would be a huge improvement because bitmaps allow VERY fast intersection through the bitwise AND operation. The current intersection algorithm needs to step through each element in given postings.

I recently wrote a small program to compare postings compression on Prometheus index files: https://github.com/thanos-io/postings-analyzer. You can see that it is possible to save around ~70% in postings size using S4-BP128-D4 and ~47% using roaring bitmaps. These numbers were consistent in my tests using index files from production. In my case, this would lead to shaving about 30% of the whole index file. Of course, most notably my index files didn’t have any runs of numbers so run-length encoding wasn’t used in roaring bitmaps, and so one could argue that I don’t have a diverse data set in these tests. Perhaps there is some weird setup out there where RLE would be useful? I tried to gather sample index files on CNCF Slack to no avail – no one stepped up to upload them for me.

Either way, all of this work is very promising and I hope to have a feature flag in Thanos soon which would allow using roaring bitmaps!

Is it a good idea to use Prometheus for storing ASCII paintings?

In the Cloud-based computing world, a relatively popular free and open-source software product called Prometheus exists which lets you monitor and observe other things. One of the components of its user interface lets you execute ad-hoc queries on the data that it has and see their results – not just in a table but also in a graphical way as well. For example, this is a query time() which plots the current time using two dimensions:

So, this gave me an idea some time ago: why not try to put some ASCII paintings in that interface and see how well Prometheus would be able to store them? And that is what I have done. To test this out, I needed to create a simple HTTP server which would serve the “metrics” which are actually the painting parts.

I have done it using the Rust programming language: additionally I got some experience in dealing with HTTP requests in it since I am still new to it. Lets continue talking about the actual realization of this thing. Note: if you ever have any trouble viewing the images then please right click on them and press View Image.

Implementation detail

Downsides

Immediately, a keen reader would have noticed that you cannot completely map the original ASCII paintings to the Prometheus interface since the characters could take any of the 255 different, possible forms, and we only have lines, albeit they can be with different colors, at our disposal.

However, the colors will be the representation of newline characters in the original painting. Thus, unfortunately, the different characters will have to be transformed into either 1 or 0 or, in other words, either a dot exists – the character is not space – or not.

So, we will inevitably lose some kind of information about the painting so it is a relatively lossy encoding scheme 🙁 But even in the face of it, lets continue on with our fun experiment.

Another thing to consider is the gap between different lines. Prometheus metrics have a floating point value attached to them. We could use 1naively everywhere as the value that we will add to separate two different lines however that will not get us very far ahead since the Prometheus UI automatically adapts the zoom level and the maximum values on the X/Y axis according to the retrieved data. This means that we might still have relatively big gaps even with that.

For that reason, we need to introduce some kind of “compression factor” into our application. Using it, we would be able to “squish” the painting more or “expand” it so that it would encompass more space at the expense of prettiness and recognizability.

Keeping that in mind, lets continue on to a example painting so that we would be able to see how it looks like.

Example

Lets start with the classical Tux penguin:

            .-"""-.
           '       \
          |,.  ,-.  |
          |()L( ()| |
          |,'  `".| |
          |.___.',| `
         .j `--"' `  `.
        / '        '   \
       / /          `   `.
      / /            `    .
     / /              l   |
    . ,               |   |
    ,"`.             .|   |
 _.'   ``.          | `..-'l
|       `.`,        |      `.
|         `.    __.j         )
|__        |--""___|      ,-'
   `"--...,+""""   `._,.-'

Our Prometheus will be configured with the following configuration:

---
global:
    scrape_interval: 1s

scrape_configs:
    - job_name: 'painter'
      static_configs:
      - targets: ['localhost:1234']

I have the scrape interval smaller so that it would take less time to ingest all of the painting into Prometheus. By running cargo run -- --input ./test4 I got the following result:

Our Tux penguin in Prometheus!

Now, lets try to compare the different compression factors and see how they play out in terms of the Prometheus query user interface:

On the left-most side we see the Tux penguin with the default compression factor i.e. 1 is used as the “gap”. In the middle, the Tux penguin became bigger by using compression factor 0.5 i.e. the penguin just became much bigger! However, as you can see, it became much harder to understand that we are still looking at the penguin. Lastly, the one on the right uses compression factor 2 or, in other words, 0.5 is used as the “gap” between lines. The penguin became much more legible in this case!

Lastly, lets try some kind of very big painting to see how well it fares in such situations as well. Try to guess what is this:

Yes, it is a duck! Here is the original ASCII:

                          XXXXXXXXXXXXXXXXXXXXXXX
                     XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
                  XXXX                                XXXX
              XXXX                                        XXXX
           XXX                                                XXX
         XX                                                      XX
       XX                                                          XX
      XX                                                            XX
     XX                                                              XX
    XX                                                        X       XX
   XX                                                   XX     XX      XX
  XX                                                      XXX    XX     XX
 XX      XX   XX                                             XX         XX
 XX    XX   XX                                                 XX        XX
XX    X    X                                                    XX       XX
XX   X    X                                                               X
X   X    X                                                                X
X       X              8                                 8                X
X                       8                               8                 X
X                  8     8                             8   8              X
X                   8  8  8                           8  8   8            X
X                    8  8  8                         8  8  88             X
X                     8  8  8                       XXXX  8               X
X                      8 XXXX                       XXXXX8                X
XX                      XXXXXX                    XXXXXXXX               XX
XX                     XXXXXXXX                  XXXXXXXXXX              XX
XX                    XXXXXXXXXX                XXXXXXXXXXXX             XX
 XX                  XXXXXXXXXXXX               XXXXXXXXXXXXX           XX
  XX                 XXXXXXXXXXXXX             XXXXXXXXXXXXXX          XX
  XX                XXXXXXXXXXXXXX            XXXXXXXXXXXXXXX          XX
  XX                XXXXXXXXXXXXXX           XXXXXXXXXXXXXXXX          XX
   XX              XXXXXXXXXXXXXXX           XXXXXXXXXXXXXXXX         XX
    XX             XXXXXXXXXXXXXXX           XXXXXXX    XXXXX        XX
     XX            XXXXXXX   XXXXX           XXXXXX      XXXX       XX
     XX            XXXXXX     XXX            XXXXX       XXXX       XX
      XX           XXXXX  88  XXXX           XXXX   88   XXX       XX
      XX           XXXX  8888  XX            XXXX  8888  XXX       XX
       XX          XXXX  8888 XXX            XXXX  8888 XXX       XX
        XX         XXXXX  88 XXX              XXXX  88 XXX       XX
          XX        XXXX    XXX               XXXX    XXX       XX
           XXX       XXXXXXXXX                 XXXXXXXXX       XXX
           XX          XXXXX      XXXXXXXXXXX    XXXXX           XX
          XXX           XX    XXXX           XXX  XX             XXX
          XX                XX XXXXX          XXXXX                XX
          XX               X  XX    XXXX  XXXX  XXXX   XXXX        XX
          XX                    XXX     XX     XX   XXX    X       XX
          XX                       XXX     XXX                    XX
           XX                         XXXXX                     XXX
             XX                                               XXX
              XXXXX                                        XXXX
                   XXXXXXXXXXX                     XXXXXXXX
                              XXXX             XXXX
                                 XX           XX
                                  XX         XX
                                  XX         XX
                                  XX         XX
                                   XX       XX
                                    XX      XX
                                    XX      XX
                                 XXXXX       XXXX
                               XX                XX
                             XX X XX        XX X XX
                            XX  XX             XX XX
                           XX  XX               XX XX
                          XX   XX               XX  XX
                         XX   XX                 XX  XX
                       XX    XX                   XX  XX
                      XX    XX                    XX   XX
                     XX    XX                     XX   XX
                    XX    XX                       XX   XX
                    XX    XX                       XX   XX
                    XX  XX                         XX   XX
                    XX  XX                         XX   XX
                     XX XX                         XX  XX
                      XXXX                         XXXX
                        XX                         XX
                        XX                         XX
                        XX                         XX
                        XX                         XX
                        XX                        XX
                        XX                       XX
                         XX                     XX
                         XX                     XX
                          XX                   XX
                           XX                 XX
                           XX     XXXXX     XX
                            XX  XX     XX  X
                             X  X       X  X
         XXXXXXX             X  X       X  X
   XXXXXX       XXXX         X  X       X  X           XXXXXX
 XXX                XXXXX    X  X       X  X      XXXXX      XXXXXX
XX     XXXX              XXXXX  X       X  X  XXXX                 XXX
X    XX  XX                     X     XXX  XXX              XXX      XX
X   X  XX                       XX   XX                      XXXX     XX
X  X XX                         XX   XX                        XXXX    XX
X X X                        XXXX     XXXX                       X XX   X
XX  X                    XXXX             XXX                     X X   X
XX X                  XXX                    XXX                   X X  X
XX XX               XX                         XXX                 X X  X
     X           XXX                              XX               X X X
      XXXXXXXXXXX                                   XXXX          XXXXX
                                                        XXXXXXXXXX

Taken from ASCIIWorld.

Disk usage comparison

Lets try to compare how much it takes to store the Tux image used before, for example. Also, note that Prometheus by itself stores some “meta” metrics about its internal state such as the metric up which shows what jobs were up and if they were successfully scraped or not.

By itself the Tux painting has 464 bytes of data. I ran Prometheus again and “painted” the ASCII picture there. The end result is that for storing all of it + some meta metrics it takes 10232 bytes of disk space.

Given that it is such a lossy encoding scheme and that it takes ~25 more times to store the same picture of Tux we can safely conclude that it is not a good idea to store our paintings there.

Future

Perhaps we could take this concept even further and write a FUSE filesystem for Linux which would store all of this data in Prometheus? We have all of the needed components: we are able to store ones and zeros, and one other symbol to separate between different “parts”. Plus, this filesystem would also provide a very “futuristic” feature – we would be able to travel back in time to see how the contents of the disk have changed.

On the other hand, spurious network problems could lead to data loss since Prometheus would not be able to scrape all of the metrics. So perhaps this idea should be abandoned after all unless someone wants to do such an experiment.

You can find all of the source code here! Do not hesitate to comment or share this if you have enjoyed it.