Prometheus is an open-source systems monitoring and alerting toolkit with an active ecosystem. See the overview.
See the comparison page.
The main Prometheus server runs standalone and has no external dependencies.
Yes, run identical Prometheus servers on two or more separate machines. Identical alerts will be deduplicated by the Alertmanager.
There are in fact various ways to scale and federate Prometheus. Read Scaling and Federating Prometheus on the Robust Perception blog to get started.
Most Prometheus components are written in Go. Some are also written in Java, Python, and Ruby.
All repositories in the Prometheus GitHub organization that have reached version 1.0.0 broadly follow semantic versioning. Breaking changes are indicated by increments of the major version. Exceptions are possible for experimental components, which are clearly marked as such in announcements.
Even repositories that have not yet reached version 1.0.0 are in general quite
stable. We aim for a proper release process and an eventual 1.0.0 release for
each repository. In any case, breaking changes will be pointed out in release
notes (marked by
[CHANGE]) or communicated clearly for components that do not
have formal releases yet.
Pulling over HTTP offers a number of advantages:
Overall we believe that pulling is slightly better than pushing, but it should not be considered a major point when considering a monitoring system.
The Push vs Pull for Monitoring blog post by Brian Brazil goes into more detail.
For cases where you must push, we offer the Pushgateway.
Short answer: Don't! Use something like the ELK stack instead.
Longer answer: Prometheus is a system to collect and process metrics, not an event logging system. The Raintank blog post Logs and Metrics and Graphs, Oh My! provides more details about the differences between logs and metrics.
If you want to extract Prometheus metrics from application logs, Google's mtail might be helpful.
It's now maintained and extended by a wide range of companies and individuals.
Prometheus is released under the Apache 2.0 license.
After extensive research, it has been determined that the correct plural of 'Prometheus' is 'Prometheis'.
Yes, sending SIGHUP to the Prometheus process or an HTTP POST request to the
/-/reload endpoint will reload and apply the configuration file. The
various components attempt to handle failing changes gracefully.
Yes, with the Alertmanager.
Currently, the following external systems are supported:
To avoid any kind of timezone confusion, especially when the so-called daylight saving time is involved, we decided to exclusively use Unix time internally and UTC for display purposes in all components of Prometheus. A carefully done timezone selection could be introduced into the UI. Contributions are welcome. See issue #500 for the current state of this effort.
There are a number of client libraries for instrumenting your services with Prometheus metrics. See the client libraries documentation for details.
If you are interested in contributing a client library for a new language, see the exposition formats.
Yes, the Node Exporter exposes an extensive set of machine-level metrics on Linux and other Unix systems such as CPU usage, memory, disk utilization, filesystem fullness and network bandwidth.
Yes, the SNMP Exporter allows monitoring of devices that support SNMP.
Yes, for applications that you cannot instrument directly with the Java client you can use the JMX Exporter either standalone or as a Java Agent.
Performance across client libraries and languages may vary. For Java, benchmarks indicate that incrementing a counter/gauge with the Java client will take 12-17ns, depending on contention. This is negligible for all but the most latency-critical code.
You are suffering from an unclean shutdown. Prometheus has to shut down cleanly
SIGTERM, which might take a while for heavily used servers. If the
server crashes or is killed hard (e.g. OOM kill by the kernel or your runlevel
system got impatient while waiting for Prometheus to shutdown), a crash
recovery has to be performed, which should take less than a minute under normal
circumstances, but can take quite long under certain circumstances. See
crash recovery for details.
See the section about memory usage to configure Prometheus for the amount of memory you have available.
Your storage is under heavy load. Read the section about configuring the local storage to find out how you can tweak settings for better performance.
We restrained ourselves to 64-bit floats to simplify the design. The IEEE 754 double-precision binary floating-point format supports integer precision for values up to 253. Supporting native 64 bit integers would (only) help if you need integer precision above 253 but below 263. In principle, support for different sample value types (including some kind of big integer, supporting even more than 64 bit) could be implemented, but it is not a priority right now. Note that a counter, even if incremented one million times per second, will only run into precision issues after over 285 years.
Initially, Prometheus ran completely on LevelDB, but to achieve better performance, we had to change the storage for bulk sample data. We evaluated many storage backends that were available at the time, without getting satisfactory results. So we implemented exactly the parts we needed, while keeping LevelDB for indexes and making heavy use of file system capabilities. Obviously, we could not evaluate every single storage backend out there, and storage backends have evolved meanwhile. However, the performance of the solution implemented now is satisfactory for most use-cases. Our most important requirements are an acceptable query speed for common queries and a sustainable ingestion rate of hundreds of thousands of samples per second. The latter depends on many parameters, like the compressibility of the sample data, the number of time series the samples belong to, the retention policy, and even more subtle aspects like how full your SSD is. If you want to know all the details, read this document with detailed benchmark results. The highlights:
On a typical bare-metal server with 64GiB RAM, 32 CPU cores, and SSD, Prometheus sustained an ingestion rate of 900k samples per second, belonging to 1M time series, scraped from 720 targets.
On a server with HDD and 128GiB RAM, Prometheus sustained an ingestion rate of 250k samples per second, belonging to 1M time series, scraped from 720 targets.
Running out of inodes is unlikely in a usual set-up. However, if you have a lot of short-lived time series, or you have configured your file system with an unusual low amount of inodes, you might run into inode depletion. Also, if you want to delete Prometheus's storage directory, you will notice that some file systems are very slow when deleting a large number of files.
While TLS and authentication are frequently requested features, we have intentionally not implemented them in any of Prometheus's server-side components. There are so many different options and parameters for both (10+ options for TLS alone) that we have decided to focus on building the best monitoring system possible rather than supporting fully generic TLS and authentication solutions in every server component.
If you need TLS or authentication, we recommend putting a reverse proxy in front of Prometheus. See for example Adding Basic Auth to Prometheus with Nginx.
Note that this applies only to inbound connections. Prometheus does support scraping TLS- and auth-enabled targets, and other Prometheus components that create outbound connections have similar support.