Prometheus includes a local on-disk time series database, but also optionally integrates with remote storage systems.
Prometheus's local time series database stores data in a custom, highly efficient format on local storage.
Ingested samples are grouped into blocks of two hours. Each two-hour block consists of a directory containing a chunks subdirectory containing all the time series samples for that window of time, a metadata file, and an index file (which indexes metric names and labels to time series in the chunks directory). The samples in the chunks directory are grouped together into one or more segment files of up to 512MB each by default. When series are deleted via the API, deletion records are stored in separate tombstone files (instead of deleting the data immediately from the chunk segments).
The current block for incoming samples is kept in memory and is not fully
persisted. It is secured against crashes by a write-ahead log (WAL) that can be
replayed when the Prometheus server restarts. Write-ahead log files are stored
wal directory in 128MB segments. These files contain raw data that
has not yet been compacted; thus they are significantly larger than regular block
files. Prometheus will retain a minimum of three write-ahead log files.
High-traffic servers may retain more than three WAL files in order to keep at
least two hours of raw data.
A Prometheus server's data directory looks something like this:
│ └── meta.json
│ ├── chunks
│ │ └── 000001
│ ├── tombstones
│ ├── index
│ └── meta.json
│ └── meta.json
│ ├── chunks
│ │ └── 000001
│ ├── tombstones
│ ├── index
│ └── meta.json
│ └── 000001
Note that a limitation of local storage is that it is not clustered or replicated. Thus, it is not arbitrarily scalable or durable in the face of drive or node outages and should be managed like any other single node database. The use of RAID is suggested for storage availability, and snapshots are recommended for backups. With proper architecture, it is possible to retain years of data in local storage.
Alternatively, external storage may be used via the remote read/write APIs. Careful evaluation is required for these systems as they vary greatly in durability, performance, and efficiency.
For further details on file format, see TSDB format.
The initial two-hour blocks are eventually compacted into longer blocks in the background.
Compaction will create larger blocks containing data spanning up to 10% of the retention time, or 31 days, whichever is smaller.
Prometheus has several flags that configure local storage. The most important are:
--storage.tsdb.path: Where Prometheus writes its database. Defaults to
--storage.tsdb.retention.time: When to remove old data. Defaults to
storage.tsdb.retention if this flag is set to anything other than default.
--storage.tsdb.retention.size: The maximum number of bytes of storage blocks to retain.
The oldest data will be removed first. Defaults to
0 or disabled. Units supported:
B, KB, MB, GB, TB, PB, EB. Ex: "512MB". Based on powers-of-2, so 1KB is 1024B. Only
the persistent blocks are deleted to honor this retention although WAL and m-mapped
chunks are counted in the total size. So the minimum requirement for the disk is the
peak space taken by the
wal (the WAL and Checkpoint) and
(m-mapped Head chunks) directory combined (peaks every 2 hours).
--storage.tsdb.retention: Deprecated in favor of
--storage.tsdb.wal-compression: Enables compression of the write-ahead log (WAL).
Depending on your data, you can expect the WAL size to be halved with little extra
cpu load. This flag was introduced in 2.11.0 and enabled by default in 2.20.0.
Note that once enabled, downgrading Prometheus to a version below 2.11.0 will
require deleting the WAL.
Prometheus stores an average of only 1-2 bytes per sample. Thus, to plan the capacity of a Prometheus server, you can use the rough formula:
needed_disk_space = retention_time_seconds * ingested_samples_per_second * bytes_per_sample
To lower the rate of ingested samples, you can either reduce the number of time series you scrape (fewer targets or fewer series per target), or you can increase the scrape interval. However, reducing the number of series is likely more effective, due to compression of samples within a series.
If your local storage becomes corrupted for whatever reason, the best strategy to address the problem is to shut down Prometheus then remove the entire storage directory. You can also try removing individual block directories, or the WAL directory to resolve the problem. Note that this means losing approximately two hours data per block directory. Again, Prometheus's local storage is not intended to be durable long-term storage; external solutions offer extended retention and data durability.
If both time and size retention policies are specified, whichever triggers first will be used.
Expired block cleanup happens in the background. It may take up to two hours to remove expired blocks. Blocks must be fully expired before they are removed.
Prometheus's local storage is limited to a single node's scalability and durability. Instead of trying to solve clustered storage in Prometheus itself, Prometheus offers a set of interfaces that allow integrating with remote storage systems.
Prometheus integrates with remote storage systems in three ways:
The read and write protocols both use a snappy-compressed protocol buffer encoding over HTTP. The protocols are not considered as stable APIs yet and may change to use gRPC over HTTP/2 in the future, when all hops between Prometheus and the remote storage can safely be assumed to support HTTP/2.
The built-in remote write receiver can be enabled by setting the
--web.enable-remote-write-receiver command line flag. When enabled,
the remote write receiver endpoint is
For details on the request and response messages, see the remote storage protocol buffer definitions.
Note that on the read path, Prometheus only fetches raw series data for a set of label selectors and time ranges from the remote end. All PromQL evaluation on the raw data still happens in Prometheus itself. This means that remote read queries have some scalability limit, since all necessary data needs to be loaded into the querying Prometheus server first and then processed there. However, supporting fully distributed evaluation of PromQL was deemed infeasible for the time being.
To learn more about existing integrations with remote storage systems, see the Integrations documentation.
If a user wants to create blocks into the TSDB from data that is in OpenMetrics format, they can do so using backfilling. However, they should be careful and note that it is not safe to backfill data from the last 3 hours (the current head block) as this time range may overlap with the current head block Prometheus is still mutating. Backfilling will create new TSDB blocks, each containing two hours of metrics data. This limits the memory requirements of block creation. Compacting the two hour blocks into larger blocks is later done by the Prometheus server itself.
A typical use case is to migrate metrics data from a different monitoring system or time-series database to Prometheus. To do so, the user must first convert the source data into OpenMetrics format, which is the input format for the backfilling as described below.
Backfilling can be used via the Promtool command line. Promtool will write the blocks to a directory. By default this output directory is ./data/, you can change it by using the name of the desired output directory as an optional argument in the sub-command.
promtool tsdb create-blocks-from openmetrics <input file> [<output directory>]
After the creation of the blocks, move it to the data directory of Prometheus.
If there is an overlap with the existing blocks in Prometheus, the flag
--storage.tsdb.allow-overlapping-blocks needs to be set for Prometheus versions
v2.38 and below. Note that any backfilled data is subject to the retention
configured for your Prometheus server (by time or size).
By default, the promtool will use the default block duration (2h) for the blocks; this behavior is the most generally applicable and correct. However, when backfilling data over a long range of times, it may be advantageous to use a larger value for the block duration to backfill faster and prevent additional compactions by TSDB later.
--max-block-duration flag allows the user to configure a maximum duration of blocks.
The backfilling tool will pick a suitable block duration no larger than this.
While larger blocks may improve the performance of backfilling large datasets, drawbacks exist as well. Time-based retention policies must keep the entire block around if even one sample of the (potentially large) block is still within the retention policy. Conversely, size-based retention policies will remove the entire block even if the TSDB only goes over the size limit in a minor way.
Therefore, backfilling with few blocks, thereby choosing a larger block duration, must be done with care and is not recommended for any production instances.
When a new recording rule is created, there is no historical data for it.
Recording rule data only exists from the creation time on.
promtool makes it possible to create historical recording rule data.
To see all options, use:
$ promtool tsdb create-blocks-from rules --help.
$ promtool tsdb create-blocks-from rules \
--start 1617079873 \
--end 1617097873 \
--url http://mypromserver.com:9090 \
The recording rule files provided should be a normal Prometheus rules file.
The output of
promtool tsdb create-blocks-from rules command is a directory that
contains blocks with the historical rule data for all rules in the recording rule
files. By default, the output directory is
data/. In order to make use of this
new block data, the blocks must be moved to a running Prometheus instance data dir
storage.tsdb.path (for Prometheus versions v2.38 and below, the flag
--storage.tsdb.allow-overlapping-blocks must be enabled). Once moved, the new
blocks will merge with existing blocks when the next compaction runs.
interval is set in the recording rule file that will take priority over
eval-interval flag in the rule backfill command.
This documentation is open-source. Please help improve it by filing issues or pull requests.