Aliaksandr Valialkin
@valyala
Followers
1,117
Following
628
Media
5
Statuses
1,163
Explore trending content on Musk Viewer
#PressConMGSongKhla25xENGLOT
• 364149 Tweets
#LingOrmFMinMNL
• 351807 Tweets
LINGORM MANILA FANMEET
• 295971 Tweets
Miami
• 102686 Tweets
FOURTH SHINING IN KOREA
• 87049 Tweets
CWR x RAVIPA Jewelry
• 83121 Tweets
石破首相
• 67845 Tweets
#スパスタ3期
• 62726 Tweets
#Liella
• 58396 Tweets
ラブライブ
• 52308 Tweets
スーパースター
• 46704 Tweets
LUX PPNARAVIT
• 37332 Tweets
凱旋門賞
• 36364 Tweets
YabaniDizisi DevamEtsin
• 33913 Tweets
京都大賞典
• 29022 Tweets
ディープボンド
• 27091 Tweets
ブローザホーン
• 21198 Tweets
FORTPEAT LOVESEA FANMEETING
• 18109 Tweets
ドラゴンズ
• 17743 Tweets
ビシエド
• 16361 Tweets
立浪監督
• 14780 Tweets
シックスペンス
• 11530 Tweets
WIN5
• 11062 Tweets
Tried to explain why OpenTelemetry format for metrics is bloated and over-engineered, and why Prometheus remote_write format is better -
6
20
117
Why everybody wants HA (high availability) and nobody mentions DR (disaster recovery)? When HA fails (it fails most of the time because of inherent complexity and fragility - just google for outages in known companies), then DR should come to rescue.
5
9
98
Check out - simple building blocks for programmatic marshaling and unmarshaling or arbitrary protobuf messages in
@golang
!
4
17
95
Very detailed article from
@zomato
about their observability journey from Thanos to VictoriaMetrics -
0
7
65
Did you know that the performance of modern computers with tens or hundreds of CPU cores is limited by memory bandwidth? E.g. prefer reducing working set, so it fits CPU caches, instead of optimizing for CPU instructions.
2
4
60
Published an article on how VictoriaMetrics uses f-tests instead of table-driven tests in Go -
3
10
50
VictoriaMetrics reduces memory usage by up to 5x starting from v1.92.0. Below is a an example memory usage reduction after the migration from v1.91 to v1.92 in highly loaded production setup with high churn rate.
Share your results in comments!
2
18
49
How VictoriaMetrics helped Grammarly reducing their observability costs -
#graphite
#observability
#monitoring
1
12
44
A trillion of rows? What's a tiny number! That's lower than the average number of rows single-node
@VictoriaMetrics
stores in a typical production setup. How about 10 trillion of rows in single-node setup or hundreds of trillions of rows in cluster setup?
So you worry about your database at millions or billions of records in a table?
Some of our production services use
@TimescaleDB
with more than a trillion rows in a single (hyper)table.
Sharing more soon. T-10.
2
4
17
4
4
36
I'm going to uncover some details about our upcoming open source log management system - VictoriaLogs. Curios how does it compare to ElasticSearch and Grafana Loki? Then reserve a time slot in your calendar for
@VictoriaMetrics
virtual meetup on March 30!
Hear about the latest updates on VictoriaMetrics by joining us for our first virtual user meet up of the year!
Follow us on our YouTube channel & tune in live on March 30th! We look forward to seeing you there 😎
#opensource
#community
#monitoring
0
1
5
6
10
31
We at VictoriaMetrics avoid SDKs because of bloat. For example, vmagent doesn't use Kubernetes SDK, AWS SDK and GCE SDK for service discovery because they increase vmagent binary size by more than 100%. As a bonus, direct REST API usage improves debuggability and maintainability
Am I the only one who tries as hard as possible to avoid SDKs?
If I'm just going to hit a handful of your APIs, I prefer to hit the REST endpoints directly, thank you. Drastically simplifies things in these situations.
135
20
794
1
1
29
OpenTelemetry for logs looks much better than OpenTelemetry for metrics! The `resource labels` ideally fits `stream labels` concept in VictoriaLogs. We are going to add support for OTEL for logs into VictoriaLogs soon.
1
2
27
vmagent allows reducing RAM usage by more than 10x comparing to Grafana agent or Prometheus in agent mode when scraping many targets. It also reduces network bandwidth usage (and costs) by 4x!
New blog post 😎
Comparing Performance & Resource Usage:
Grafana Agent vs. Prometheus Agent Mode vs. VictoriaMetrics vmagent
Read it here:
#prometheus
#grafana
#monitoring
#observability
2
10
37
0
3
27
Protobuf compiler for
@golang
adds useless dependencies, which increase binary size by 9MB without giving any benefit. It would be great if this could fixed in the future. In the mean time we have to manually remove these dependencies.
0
2
27
@samokhvalov
@PostgreSQL
PostgreSQL may be good fit for real-time analytics over small-to-medium volumes of data (e.g. up to a few GBs and a few billions of rows). For bigger volumes I'd recommend taking a look at
@ClickHouseDB
5
5
26
Want to know more on the history of VictoriaMetrics and on internal architecture design decisions? Then listen to this podcast!
Excited to release episode
#61
on The GeekNarrator podcast. This time we dive deep into how
@VictoriaMetrics
works and provides a simple and reliable solution for monitoring at massive scale with
@hagen1778
and
@valyala
.
Along with VictoriaMetrics internals, it is also packed
1
5
34
2
4
25
The work on easyproto resulted in two performance-related proposals for
@golang
:
1.
2.
I hope they will be accepted and implemented soon.
1
1
24
What a great new year present!
More great news before the end of the year:
Happy 10K+ stars on GitHub, Vicky 🎉
Thanks for your support everyone!
0
3
11
1
3
24
The next release of VictoriaMetrics will allow reducing network costs by 2-4 times when transferring data from vmagent to VictoriaMetrics
1
7
23
@_AkihiroSuda_
Writing protobuf marshalers and unmarshalers by hands isn't too hard when using . This will protect you from potential backdoors in the generated *.pb.go code
1
5
22
Published VictoriaMetrics project goals and non-goals -
1
0
22
@OmranJamal
@TimescaleDB
Take a look at
@ClickHouseDB
- it performs analytical queries up to 10000x faster than Postgres or TimescaleDB
0
2
21
@todaywasawesome
The issue isn't in Prometheus itself. The issue is in not so good state of Prometheus monitoring stack for Kubernetes, which collects too many never used metrics with default configs. The solution is to drop such metrics with relabeling. Or to try VictoriaMetrics :)
2
2
20
Simple Prometheus text exposition format is de-facto standard for metrics exposition - it is supported by almost any application out there. Why do we need over-engineered OpenTelemetry protocol for metrics instead?
#prometheus
#opentelemetry
#cncf
1
2
21
@chinmay185
Side note: OpenTelemetry is so over-engineered and buggy, that even the best developers and SREs cannot figure out how to work properly with it :)
4
4
21
Because product managers believe that it is better to add a dozen of half-baked new features into the product instead of improving the usability and the efficiency of the existing features.
0
3
20
VictoriaMetrics reached 11111 stars on GitHub!
2
1
20
@michaelfreedman
@PostgreSQL
What about analytical workloads with the ingestion rate of hundreds of billions rows per day and online queries over the ingested rows?
@ClickHouseDB
fits better here than PostgreSQL or TimescaleDB -
2
2
19
Tomorrow I'll give a talk at FrOSSCon on why VictoriaLogs is the best solution for large-scale log collection and analysis in Kubernetes
🚀 Ready to get into log management?
Join our CTO Aliaksandr Valialkin
@valyala
on August 17th at 11:15 AM CET at
#FrOSCon
for insights into optimizing large-scale
#Kubernetes
clusters. 🛠️
Discover how to optimize your infrastructure and save 10x or more on costs. 📅
1
2
11
0
3
20
This is 100% truth! People who don't care about your software don't bother filing bugreports. They just stop using it after they encounter enough number of annoying bugs, and you'll never know about these bugs.
The people who file the most bugs against your software are the ones who love it the most. Cherish those people.
9
48
258
1
2
19
@golang
It is pity the harmful iterator "feature" wasn't removed from Go1.23 rc1 :(
I'm sorry, but Go is going in the wrong direction with range over functions, iterators, etc.
The core promise of the language was simplicity. Why the Go team is so eager to flush it down the drain is beyond me.
I never signed up for this crap.
125
71
847
3
1
18
It is better to improve usability for existing features instead of adding a ton of new half-baked unusable features and breaking existing features over time.
Why most software products still choose "a ton of new half-baked features" path?
3
1
15
LogsQL in VictoriaLogs becomes awesome to use with recent updates! Try it and you'll definitely fall in love with its simplicity and power!
Since the last update it gained the ability to parse and format log fields, plus the ability to use subqueries inside `in(...)` filters!
LogsQL gains convenient functionality for fast statistics' calculation over the selected logs starting from v0.6.0! Also it returns all the fields for the selected logs by default - this significantly simplifies logs' investigation.
0
1
12
0
0
16
Which practical problem does this feature resolve? I see only theoretical "problems". And real problems with the complexity this feature adds to Go.
4
0
17
The recent release of VictoriaMetrics (v1.87.1 - ) has been built with Go1.20. It shows up to 5% performance improvements in data ingestion benchmarks comparing to the previous release built with Go1.19.
Nice improvement by just upgrading Go builder!
0
0
17
Good to see that Prometheus adopts usability features from VictoriaMetrics! This feature has been added to VictoriaMetrics 3 years ago -
It would be great to see the ability to use K, M, G, T, KB, MB, GB, KiB MiB, GiB and TiB suffixes in numeric constants
Version 2.54 of
@PrometheusIO
comes with a nice new PromQL usability feature that allows you to use duration and number literals interchangeably, so you can write queries like this:
🔍 time() - my_timestamp_seconds > 30d
I wrote a blog post about it!
1
3
15
0
1
17
It looks like Go transforms from community-first to enterprise-first project internal Google project with enterprise goals, processes and priorities, according to this podcast :(
This explains the direction towards increased complexity
🚀 New episode of Go Time!
💬 How things get done on the Go Team
🌟 with
@cameronbalahan
@Sajma
&
@_rsc
🎙 hosted by
@Angelica_Hill
🎧
#golang
1
10
22
10
1
16
0
7
16
VictoriaLogs can accept logs via Syslog protocol via UDP, TCP and TLS+TCP starting from v0.20.1 release!
0
4
16
This is a task, which should be targeted by Go core team, instead of adding generics, iterators and other useless "features"
1
3
14
This can be also a big issue for Kubernetes nodes with big number of CPU cores. A pod, which eats all the memory bandwidth on the node, can significantly slow down other pods, which run on the same node, even if there is a plenty of free CPU there.
2
3
15
If you are stuck with the decision because multiple decision options exist and it is unclear which is better, then stop wasting your time and choose a random option. This strategy works good everywhere: in software development, in management, in personal relationship, etc.
2
0
15
@go100and1
That's horribly complex article! I'd expect such an article from C++ developers. Unfortunately, generics irreversibly complicated Go ecosystem :(
3
1
14
@MattJamesBoyle
@VictoriaMetrics
There is also an elegant outliers' detection technique based on interquartile range (IQR). It may work better than z-score based anomaly detection in some cases. The next release of VictoriaMetrics will provide functions for detecting anomalies via IQR.
4
2
14
...and why Go programmers prefer catching all the panics at the top level of HTTP handlers instead of allowing Go runtime to crash with the panic stack trace? This hides the root cause of the bug, which triggers the panic. This also may leave the app in inconsistent state.
4
1
13
VictoriaLogs web UI gains the ability to show bar chart with the number matching logs on the selected time range. This helps logs' investigation -
0
2
13
VictoriaLogs just gained the ability to return surrounding logs in front and after the selected logs -
0
0
14
VictoriaMetrics speeds up repeated SLO / SLI alerting queries over long lookbehind windows starting from v1.95.1 . For example, `avg_over_time(up[30d]) < 0.99` should be performed much faster now.
2
0
14
@rakyll
While PostgreSQL is great, it is much better to use specialized databases for big volumes of time series data -
@ClickHouseDB
and
@VictoriaMetrics
. They can store and process multiple orders of magnitude more data than PostgreSQL on the same hardware at much higher speed.
4
0
14
VictoriaLogs preview is out! Try it on your production workloads and feel the difference with Elasticsearch and Grafana Loki
Key VictoriaLogs Highlights 😎
🔹Requires up to 10x less disk space & RAM on production workloads
🔹Easier to configure & operate than similar solutions
🔹LogsQL: Simple, yet powerful query language
Find out more:
#opensource
#monitoring
#observability
2
5
10
0
4
12
That's an excellent example on how product features should evolve!
For anyone who wants to try out the new
@PrometheusIO
UI (>90% done) that I'm currently building for Prometheus 3.0:
⬇️ git clone git
@github
.com:prometheus/prometheus
➡️ git checkout mantine-ui
🔁 make build
▶️ ./prometheus --enable-feature=new-ui
Happy to hear feedback :)
5
9
65
0
0
13
We store vendor directory in VictoriaMetrics source code. This has additional benefits on top of those mentioned by Matt:
- simplified `git grep` and `vim+ctags` over third-party dependencies
- easier security audit of third-party code, especially during version upgrades
You should probably be committing your Vendor folder
Here's why:
- You can build your code without having to pull from external sources. This can be very important for CI pipelines when you do not have internet access (which is usually good practise).
- Everything needed to
16
8
54
1
2
13
If somebody is looking on how to start contributing to
@VictoriaMetrics
community, then follow the list below!
Our slack is at
Some ways to get started contributing to open-source (in general, every project and maintainer has differing views on value):
* Join their community!! (slack, mailing list, IRC, whatever)
* Read the docs(!!!!)
- Fix typos/grammaros/broken links
- Ask questions in the
1
10
55
1
3
13
This is so true!
Half of the job as a senior engineer is to stop team members from overengineering stuff.
Throughout my career, I have consulted at many companies where the team wanted to:
- rewrite the web API in a different programming language instead of fixing the underlying bad DB queries
17
23
310
0
0
12
Custom fonts do not bring any benefits to user experience. Default system fonts have UX benefits:
- Consistency across different tools in the given environment
- Default fonts are optimized for readability, correctness and performance
We have 100 performance score and 400ms DOMContentLoaded in VictoriaMetrics UI.
But
@valyala
is still not happy and wants to remove custom fonts to make it more lightweight :) Try our playground here
#victoriametrics
0
2
10
2
0
12
The next release of VictoriaMetrics and vmagent will support streaming aggregation!
0
5
12
@0xglitchbyte
Only 2 programming languages should exist
C for low-level stuff and Go (without generics) for anything else
every other language is bloat
0
1
12
LogsQL gains convenient functionality for fast statistics' calculation over the selected logs starting from v0.6.0! Also it returns all the fields for the selected logs by default - this significantly simplifies logs' investigation.
0
1
12
@eatonphil
Probably, they realized that the initial license change was wrong from business perspective.
1
1
12
@Cloudflare
I work right now on VictoriaLogs - open-source killer for Grafana Loki and for ELK stack (eg. Elasticsearch for logs)
1
0
12
@rakyll
Some people want distributed systems, which use sophisticated algorithms for automatic scaling and healing. Then they realize that it is better to have boring system with simple easy-to-debug architecture after a major incident with sophisticated system.
0
2
10
Unpopular opinion: OpenTelemetry is modern SOAP ( )
6
1
10
@techsavvytravvy
@ohmypy
You can't avoid garbage features in programming languages, since they will be used by others. Eventually you'll need to deal with overcomplicated shitty code, which uses these "features".
1
0
12
@wonydev
Yes, we are looking for Go software engineers with experience in observability solutions such as Prometheus and VictoriaMetrics! Please drop you CV to info
@victoriametrics
.com
1
3
11
@awesomekling
If the PR is useful (e.g. if fixes some bug or adds some useful feature), but doesn't meet some code standards used the project, then I merge the PR and then refactor it in the follow-up commit. This is my sin, but it speeds up code development and engages community.
1
0
10
@SergioRocks
Software engineers do not change their laptops every month. For example, I still use a laptop from 2018. So the laptop price should be irrelevant for most developers around the world. It's better to talk about cost of living in different parts of the world.
1
0
10
Unfortunately, this is common software engineering practice - to add a ton of half-baked features and to change the UI in horrible ways instead of polishing existing features and keeping consistent UI. Examples out of my head:
@googleanalytics
,
@grafana
,
@Android
,
@Windows
@sudomateo
@SlackHQ
I was a long time Rhapsody user (Now Napster) for listening to music. Their browser app at the time was really nice. Then overnight it changed and it was horrible. I reached out and asked why they did this. My answer was...
Progress
At the time I didn't get it. Today I realize
0
0
4
0
1
9
ChatGPT mimicks an average Reddit visitor :)
I trained a GPT on thousands of posts from the
@golang
subreddit and it's FLAWLESS. Just as unhelpful as the real thing.
6
9
100
1
0
10
@eatonphil
I'm working on a Go package, which will provide simple building blocks for fast zero-alloc marshaling and unmarshaling of protobuf messages with arbitrary nesting and complexity. This package will be used in
@VictoriaMetrics
source code initially.
2
2
10
Hurry up with buying tickets to PromCon - the best conference about Prometheus ecosystem!
🗓️ The PromCon 2024 schedule is out!
Get your tickets for PromCon 2024 in Berlin today! 🎫
0
15
30
0
0
10
The just released vmagent v1.97.1 should use lower amounts of RAM and CPU when scraping large responses from targets such as kube-state-metric!
0
0
10
@realmeson10
IMHO, the following doc explains clearly the meaining of metrics, samples, time series and cardinality -
1
1
10
@go100and1
I figured out the real purpose of generics in
@golang
- to make possible to write such quizzes :)
2
2
10
@prathamesh2_
VictoriaMetrics also gained support for OpenTelemetry data ingestion since v1.92.0
0
0
10
Interesting approach for protobuf parsing / marshaling in
@golang
It would be great if it will be eventually included into the official protobuf package
1
0
10
@heyandras
1. Easy to install agent, which automatically discovers and collects logs and metrics from popular applications at the host it runs on, without any configuration.
2. Well-thought dashboards and alerts for the collected metrics and logs, which are provided out of the box.
That's
1
1
10
What's new has been appeared in VictoriaLogs during the last 2 months -
1
0
10
I disagree. Faster computers for developers lead to slower builds, slower tests and slower bloated programs for the end users. It is better to spend developer time on optimizing slow builds, tests and the product for end users than giving them faster computers.
It’s amazing how often companies waste very expensive developer time because they are too cheap to buy them fast computers.
52
49
1K
2
0
8