Storage

Prometheus has a sophisticated local storage subsystem. For indexes, it uses LevelDB. For the bulk sample data, it has its own custom storage layer, which organizes sample data in chunks of constant size (1024 bytes payload). These chunks are then stored on disk in one file per time series.

This sections deals with the various configuration settings and issues you might run into. To dive deeper into the topic, check out the following talks: * The Prometheus Time Series Database. * Configuring Prometheus for High Performance.

Memory usage

Prometheus keeps all the currently used chunks in memory. In addition, it keeps as many most recently used chunks in memory as possible. You have to tell Prometheus how much memory it may use for this caching. The flag storage.local.target-heap-size allows you to set the heap size (in bytes) Prometheus aims not to exceed. Note that the amount of physical memory the Prometheus server will use is the result of complex interactions of the Go runtime and the operating system and very hard to predict precisely. As a rule of thumb, you should have at least 50% headroom in physical memory over the configured heap size. (Or, in other words, set storage.local.target-heap-size to a value of two thirds of the physical memory limit Prometheus should not exceed.)

The default value of storage.local.target-heap-size is 2GiB and thus tailored to 3GiB of physical memory usage. If you have less physical memory available, you have to lower the flag value. If you have more memory available, you should raise the value accordingly. Otherwise, Prometheus will not make use of the memory and thus will perform much worse than it could.

Because Prometheus uses most of its heap for long-lived allocations of memory chunks, the garbage collection target percentage is set to 40 by default. You can still override this setting via the GOGC environment variable as usual.

For high-performance set-ups, you might need to adjust more flags. Please read through the sections below for details.

Prior to v1.6, there was no flag storage.local.target-heap-size. Instead, the number of chunks kept in memory had to be configured using the flags storage.local.memory-chunks and storage.local.max-chunks-to-persist. These flags still exist for compatibility reasons. However, storage.local.max-chunks-to-persist has no effect anymore, and if storage.local.memory-chunks is set to a non-zero value x, it is used to override the value for storage.local.target-heap-size to 3072*x.

Disk usage

Prometheus stores its on-disk time series data under the directory specified by the flag storage.local.path. The default path is ./data (relative to the working directory), which is good to try something out quickly but most likely not what you want for actual operations. The flag storage.local.retention allows you to configure the retention time for samples. Adjust it to your needs and your available disk space.

Chunk encoding

Prometheus currently offers three different types of chunk encodings. The chunk encoding for newly created chunks is determined by the -storage.local.chunk-encoding-version flag. The valid values are 0, 1, or 2.

Type 0 is the simple delta encoding implemented for Prometheus's first chunked storage layer. Type 1 is the current default encoding, a double-delta encoding with much better compression behavior than type 0. Both encodings feature a fixed byte width per sample over the whole chunk, which allows fast random access. While type 0 is the fastest encoding, the difference in encoding cost compared to encoding 1 is tiny. Due to the better compression behavior of type 1, there is really no reason to select type 0 except compatibility with very old Prometheus versions.

Type 2 is a variable bit-width encoding, i.e. each sample in the chunk can use a different number of bits. Timestamps are double-delta encoded, too, but with a slightly different algorithm. A number of different encoding schemes are available for sample values. The choice is made per chunk based on the nature of the sample values (constant, integer, regularly increasing, random…). Major parts of the type 2 encoding are inspired by a paper published by Facebook engineers: Gorilla: A Fast, Scalable, In-Memory Time Series Database.

With type 2, access within a chunk has to happen sequentially, and the encoding and decoding cost is a bit higher. Overall, type 2 will cause more CPU usage and increased query latency compared to type 1 but offers a much improved compression ratio. The exact numbers depend heavily on the data set and the kind of queries. Below are results from a typical production server with a fairly expensive set of recording rules.

Chunk type bytes per sample cores rule evaluation duration
1 3.3 1.6 2.9s
2 1.3 2.4 4.9s

You can change the chunk encoding each time you start the server, so experimenting with your own use case is encouraged. Take into account, however, that only newly created chunks will use the newly selected chunk encoding, so it will take a while until you see the effects.

For more details about the trade-off between the chunk encodings, see this blog post.

Settings for high numbers of time series

Prometheus can handle millions of time series. However, with the above mentioned default setting for storage.local.target-heap-size, you will be limited to about 200,000 time series simultaneously present in memory. For more series, you need more memory, and you need to configure Prometheus to make use of it as described above.

Each of the aforementioned chunks contains samples of a single time series. A time series is thus represented as a series of chunks, which ultimately end up in a time series file (one file per time series) on disk.

A series that has recently received new samples will have an open incomplete head chunk. Once that chunk is completely filled, or the series hasn't received samples in a while, the head chunk is closed and becomes a chunk waiting to be appended to its corresponding series file, i.e. it is waiting for persistence. After the chunk has been persisted to disk, it becomes evictable, provided it is not currently used by a query. Prometheus will evict evictable chunks from memory to satisfy the configured target heap size. A series with an open head chunk is called an active series. This is different from a memory series, which also includes series without an open head chunk but still other chunks in memory (whether waiting for persistence, used in a query, or evictable). A series without any chunks in memory may be archived, upon which it ceases to have any mandatory memory footprint.

The amount of chunks Prometheus can keep in memory depends on the flag value for storage.local.target-heap-size and on the amount of memory used by everything else. If there are not enough chunks evictable to satisfy the target heap size, Prometheus will throttle ingestion of more samples (by skipping scrapes and rule evaluations) until the heap has shrunk enough. Throttled ingestion is really bad for various reasons. You really do not want to be in that situation.

Open head chunks, chunks still waiting for persistence, and chunks being used in a query are not evictable. Thus, the reasons for the inability to evict enough chunks include the following: 1. Queries that use too many chunks. 2. Chunks are piling up waiting for persistence because the storage layer cannot keep up writing chunks. 3. There are too many active time series, which results in too many open head chunks.

Currently, Prometheus has no defence against case (1). Abusive queries will essentially OOM the server.

To defend against case (2), there is a concept of persistence urgency explained in the next section.

Case (3) depends on the targets you monitor. To mitigate an unplanned explosion of the number of series, you can limit the number of samples per individual scrape (see sample_limit in the scrape config). If the number of active time series exceeds the number of memory chunks the Prometheus server can afford, the server will quickly throttle ingestion as described above. The only way out if this is to give Prometheus more RAM or reduce the number of time series to ingest.

In fact, you want many more memory chunks than you have series in memory. Prometheus tries to batch up disk writes as much as possible as it helps for both HDD (write as much as possible after each seek) and SSD (tiny writes create write amplification, which limits the effective throughput and burns much more quickly through the lifetime of the device). The more Prometheus can batch up writes, the more efficient is the process of persisting chunks to disk. which helps case (2).

In conclusion, to keep the Prometheus server healthy, make sure it has plenty of headroom of memory chunks available for the number of memory series. A factor of three is a good starting point. Refer to the section about helpful metrics to find out what to look for. A very broad rule of thumb for an upper limit of memory series is the total available physical memory divided by 10,000, e.g. About 6M memory series on a 64GiB server.

If you combine a high number of time series with very fast and/or large scrapes, the number of pre-allocated mutexes for series locking might not be sufficient. If you see scrape hiccups while Prometheus is writing a checkpoint or processing expensive queries, try increasing the value of the storage.local.num-fingerprint-mutexes flag. Sometimes tens of thousands or even more are required.

PromQL queries that involve a high number of time series will make heavy use of the LevelDB-backed indexes. If you need to run queries of that kind, tweaking the index cache sizes might be required. The following flags are relevant:

  • -storage.local.index-cache-size.label-name-to-label-values: For regular expression matching.
  • -storage.local.index-cache-size.label-pair-to-fingerprints: Increase the size if a large number of time series share the same label pair or name.
  • -storage.local.index-cache-size.fingerprint-to-metric and -storage.local.index-cache-size.fingerprint-to-timerange: Increase the size if you have a large number of archived time series, i.e. series that have not received samples in a while but are still not old enough to be purged completely.

You have to experiment with the flag values to find out what helps. If a query touches 100,000+ time series, hundreds of MiB might be reasonable. If you have plenty of memory available, using more of it for LevelDB cannot harm. More memory for LevelDB will effectively reduce the number of memory chunks Prometheus can afford.

Persistence urgency and “rushed mode”

Naively, Prometheus would all the time try to persist completed chunk to disk as soon as possible. Such a strategy would lead to many tiny write operations, using up most of the I/O bandwidth and keeping the server quite busy. Spinning disks will appear to be very slow because of the many slow seeks required, and SSDs will suffer from write amplification. Prometheus tries instead to batch up write operations as much as possible, which works better if it is allowed to use more memory.

Prometheus will also sync series files after each write (with storage.local.series-sync-strategy=adaptive, which is the default) and use the disk bandwidth for more frequent checkpoints (based on the count of “dirty series”, see below), both attempting to minimize data loss in case of a crash.

But what to do if the number of chunks waiting for persistence grows too much? Prometheus calculates a score for urgency to persist chunks. The score is between 0 and 1, where 1 corresponds to the highest urgency. Depending on the score, Prometheus will write to disk more frequently. Should the score ever pass the threshold of 0.8, Prometheus enters “rushed mode” (which you can see in the logs). In rushed mode, the following strategies are applied to speed up persisting chunks:

  • Series files are not synced after write operations anymore (making better use of the OS's page cache at the price of an increased risk of losing data in case of a server crash – this behavior can be overridden with the flag storage.local.series-sync-strategy).
  • Checkpoints are only created as often as configured via the storage.local.checkpoint-interval flag (freeing more disk bandwidth for persisting chunks at the price of more data loss in case of a crash and an increased time to run the subsequent crash recovery).
  • Write operations to persist chunks are not throttled anymore and performed as fast as possible.

Prometheus leaves rushed mode once the score has dropped below 0.7.

Throttling of ingestion happens if the urgency score reaches 1. Thus, the rushed mode is not per se something to be avoided. It is, on the contrary, a measure the Prometheus server takes to avoid the really bad situation of throttled ingestion. Occasionally entering rushed mode is OK, if it helps and ultimately leads to leaving rushed mode again. If rushed mode is entered but the urgency score still goes up, the server has a real problem.

Settings for very long retention time

If you have set a very long retention time via the storage.local.retention flag (more than a month), you might want to increase the flag value storage.local.series-file-shrink-ratio.

Whenever Prometheus needs to cut off some chunks from the beginning of a series file, it will simply rewrite the whole file. (Some file systems support “head truncation”, which Prometheus currently does not use for several reasons.) To not rewrite a very large series file to get rid of very few chunks, the rewrite only happens if at least 10% of the chunks in the series file are removed. This value can be changed via the mentioned storage.local.series-file-shrink-ratio flag. If you have a lot of disk space but want to minimize rewrites (at the cost of wasted disk space), increase the flag value to higher values, e.g. 0.3 for 30% of required chunk removal.

Crash recovery

Prometheus saves chunks to disk as soon as possible after they are complete. Incomplete chunks are saved to disk during regular checkpoints. You can configure the checkpoint interval with the flag storage.local.checkpoint-interval. Prometheus creates checkpoints more frequently than that if too many time series are in a “dirty” state, i.e. their current incomplete head chunk is not the one that is contained in the most recent checkpoint. This limit is configurable via the storage.local.checkpoint-dirty-series-limit flag.

More active time series to cycle through lead in general to more chunks waiting for persistence, which in turns leads to larger checkpoints and ultimately more time needed for checkpointing. There is a clear trade-off between limiting the loss of data in case of a crash and the ability to scale to high number of active time series. To not spend the majority of the disk throughput for checkpointing, you have to increase the checkpoint interval. Prometheus itself limits the time spent in checkpointing to 50% by waiting after each checkpoint's completion for at least as long as the previous checkpoint took.

Nevertheless, should your server crash, you might still lose data, and your storage might be left in an inconsistent state. Therefore, Prometheus performs a crash recovery after an unclean shutdown, similar to an fsck run for a file system. Details about the crash recovery are logged, so you can use it for forensics if required. Data that cannot be recovered is moved to a directory called orphaned (located under storage.local.path). Remember to delete that data if you do not need it anymore.

The crash recovery usually takes less than a minute. Should it take much longer, consult the log to find out what is going on. With increasing number of time series in the storage (archived or not), the re-indexing tends to dominate the recovery time and can take tens of minutes in extreme cases.

Data corruption

If you suspect problems caused by corruption in the database, you can enforce a crash recovery by starting the server with the flag storage.local.dirty.

If that does not help, or if you simply want to erase the existing database, you can easily start fresh by deleting the contents of the storage directory:

  1. Stop Prometheus.
  2. rm -r <storage path>/*
  3. Start Prometheus.

Helpful metrics

Out of the metrics that Prometheus exposes about itself, the following are particularly useful to tweak flags and find out about the required resources. They also help to create alerts to find out in time if a Prometheus server has problems or is out of capacity.

  • prometheus_local_storage_memory_series: The current number of series held in memory.
  • prometheus_local_storage_open_head_chunks: The number of open head chunks.
  • prometheus_local_storage_chunks_to_persist: The number of memory chunks that still need to be persisted to disk.
  • prometheus_local_storage_memory_chunks: The current number of chunks held in memory. If you substract the previous two, you get the number of persisted chunks (which are evictable if not currently in use by a query).
  • prometheus_local_storage_series_chunks_persisted: A histogram of the number of chunks persisted per batch.
  • prometheus_local_storage_persistence_urgency_score: The urgency score as discussed above.
  • prometheus_local_storage_rushed_mode is 1 if Prometheus is in “rushed mode”, 0 otherwise. Can be used to calculate the percentage of time Prometheus is in rushed mode.
  • prometheus_local_storage_checkpoint_last_duration_seconds: How long the last checkpoint took.
  • prometheus_local_storage_checkpoint_last_size_bytes: Size of the last checkpoint in bytes.
  • prometheus_local_storage_checkpointing is 1 while Prometheus is checkpointing, 0 otherwise. Can be used to calculate the percentage of time Prometheus is checkpointing.
  • prometheus_local_storage_inconsistencies_total: Counter for storage inconsistencies found. If this is greater than 0, restart the server for recovery.
  • prometheus_local_storage_persist_errors_total: Counter for persist errors.
  • prometheus_local_storage_memory_dirty_series: Current number of dirty series.
  • process_resident_memory_bytes: Broadly speaking the physical memory occupied by the Prometheus process.
  • go_memstats_alloc_bytes: Go heap size (allocated objects in use plus allocated objects not in use anymore but not yet garbage-collected).