Prometheus provides a functional query language called PromQL (Prometheus Query Language) that lets the user select and aggregate time series data in real time. The result of an expression can either be shown as a graph, viewed as tabular data in Prometheus's expression browser, or consumed by external systems via the HTTP API.
This document is meant as a reference. For learning, it might be easier to start with a couple of examples.
In Prometheus's expression language, an expression or sub-expression can evaluate to one of four types:
Depending on the use-case (e.g. when graphing vs. displaying the output of an expression), only some of these types are legal as the result from a user-specified expression. For example, an expression that returns an instant vector is the only type that can be directly graphed.
Notes about the experimental native histograms:
Strings may be specified as literals in single quotes, double quotes or backticks.
PromQL follows the same escaping rules as
Go. In single or double quotes a
backslash begins an escape sequence, which may be followed by a, b, f,
n, r, t, v or \. Specific characters can be provided using octal
(\nnn) or hexadecimal (\xnn, \unnnn and \Unnnnnnnn).
No escaping is processed inside backticks. Unlike Go, Prometheus does not discard newlines inside backticks.
Example:
"this is a string"
'these are unescaped: \n \\ \t'
`these are not unescaped: \n ' " \t`
Scalar float values can be written as literal integer or floating-point numbers in the format (whitespace only included for better readability):
[-+]?(
[0-9]*\.?[0-9]+([eE][-+]?[0-9]+)?
| 0[xX][0-9a-fA-F]+
| [nN][aA][nN]
| [iI][nN][fF]
)
Examples:
23
-2.43
3.4e-9
0x8f
-Inf
NaN
Instant vector selectors allow the selection of a set of time series and a single sample value for each at a given timestamp (instant): in the simplest form, only a metric name is specified. This results in an instant vector containing elements for all time series that have this metric name.
This example selects all time series that have the http_requests_total metric
name:
http_requests_total
It is possible to filter these time series further by appending a comma separated list of label
matchers in curly braces ({}).
This example selects only those time series with the http_requests_total
metric name that also have the job label set to prometheus and their
group label set to canary:
http_requests_total{job="prometheus",group="canary"}
It is also possible to negatively match a label value, or to match label values against regular expressions. The following label matching operators exist:
=: Select labels that are exactly equal to the provided string.!=: Select labels that are not equal to the provided string.=~: Select labels that regex-match the provided string.!~: Select labels that do not regex-match the provided string.Regex matches are fully anchored. A match of env=~"foo" is treated as env=~"^foo$".
For example, this selects all http_requests_total time series for staging,
testing, and development environments and HTTP methods other than GET.
http_requests_total{environment=~"staging|testing|development",method!="GET"}
Label matchers that match empty label values also select all time series that do not have the specific label set at all. It is possible to have multiple matchers for the same label name.
Vector selectors must either specify a name or at least one label matcher that does not match the empty string. The following expression is illegal:
{job=~".*"} # Bad!
In contrast, these expressions are valid as they both have a selector that does not match empty label values.
{job=~".+"} # Good!
{job=~".*",method="get"} # Good!
Label matchers can also be applied to metric names by matching against the internal
__name__ label. For example, the expression http_requests_total is equivalent to
{__name__="http_requests_total"}. Matchers other than = (!=, =~, !~) may also be used.
The following expression selects all metrics that have a name starting with job::
{__name__=~"job:.*"}
The metric name must not be one of the keywords bool, on, ignoring, group_left and group_right. The following expression is illegal:
on{} # Bad!
A workaround for this restriction is to use the __name__ label:
{__name__="on"} # Good!
All regular expressions in Prometheus use RE2 syntax.
Range vector literals work like instant vector literals, except that they
select a range of samples back from the current instant. Syntactically, a time
duration is appended in square brackets ([]) at the end of
a vector selector to specify how far back in time values should be fetched for
each resulting range vector element. The range is a closed interval,
i.e. samples with timestamps coinciding with either boundary of the range are
still included in the selection.
In this example, we select all the values we have recorded within the last 5
minutes for all time series that have the metric name http_requests_total and
a job label set to prometheus:
http_requests_total{job="prometheus"}[5m]
Time durations are specified as a number, followed immediately by one of the following units:
ms - millisecondss - secondsm - minutesh - hoursd - days - assuming a day has always 24hw - weeks - assuming a week has always 7dy - years - assuming a year has always 365dTime durations can be combined, by concatenation. Units must be ordered from the longest to the shortest. A given unit must only appear once in a time duration.
Here are some examples of valid time durations:
5h
1h30m
5m
10s
The offset modifier allows changing the time offset for individual
instant and range vectors in a query.
For example, the following expression returns the value of
http_requests_total 5 minutes in the past relative to the current
query evaluation time:
http_requests_total offset 5m
Note that the offset modifier always needs to follow the selector
immediately, i.e. the following would be correct:
sum(http_requests_total{method="GET"} offset 5m) // GOOD.
While the following would be incorrect:
sum(http_requests_total{method="GET"}) offset 5m // INVALID.
The same works for range vectors. This returns the 5-minute rate
that http_requests_total had a week ago:
rate(http_requests_total[5m] offset 1w)
For comparisons with temporal shifts forward in time, a negative offset can be specified:
rate(http_requests_total[5m] offset -1w)
Note that this allows a query to look ahead of its evaluation time.
The @ modifier allows changing the evaluation time for individual instant
and range vectors in a query. The time supplied to the @ modifier
is a unix timestamp and described with a float literal.
For example, the following expression returns the value of
http_requests_total at 2021-01-04T07:40:00+00:00:
http_requests_total @ 1609746000
Note that the @ modifier always needs to follow the selector
immediately, i.e. the following would be correct:
sum(http_requests_total{method="GET"} @ 1609746000) // GOOD.
While the following would be incorrect:
sum(http_requests_total{method="GET"}) @ 1609746000 // INVALID.
The same works for range vectors. This returns the 5-minute rate that
http_requests_total had at 2021-01-04T07:40:00+00:00:
rate(http_requests_total[5m] @ 1609746000)
The @ modifier supports all representation of float literals described
above within the limits of int64. It can also be used along
with the offset modifier where the offset is applied relative to the @
modifier time irrespective of which modifier is written first.
These 2 queries will produce the same result.
# offset after @
http_requests_total @ 1609746000 offset 5m
# offset before @
http_requests_total offset 5m @ 1609746000
Additionally, start() and end() can also be used as values for the @ modifier as special values.
For a range query, they resolve to the start and end of the range query respectively and remain the same for all steps.
For an instant query, start() and end() both resolve to the evaluation time.
http_requests_total @ start()
rate(http_requests_total[5m] @ end())
Note that the @ modifier allows a query to look ahead of its evaluation time.
Subquery allows you to run an instant query for a given range and resolution. The result of a subquery is a range vector.
Syntax: <instant_query> '[' <range> ':' [<resolution>] ']' [ @ <float_literal> ] [ offset <duration> ]
<resolution> is optional. Default is the global evaluation interval.Prometheus supports many binary and aggregation operators. These are described in detail in the expression language operators page.
Prometheus supports several functions to operate on data. These are described in detail in the expression language functions page.
PromQL supports line comments that start with #. Example:
# This is a comment
When queries are run, timestamps at which to sample data are selected
independently of the actual present time series data. This is mainly to support
cases like aggregation (sum, avg, and so on), where multiple aggregated
time series do not exactly align in time. Because of their independence,
Prometheus needs to assign a value at those timestamps for each relevant time
series. It does so by simply taking the newest sample before this timestamp.
If a target scrape or rule evaluation no longer returns a sample for a time series that was previously present, that time series will be marked as stale. If a target is removed, its previously returned time series will be marked as stale soon afterwards.
If a query is evaluated at a sampling timestamp after a time series is marked stale, then no value is returned for that time series. If new samples are subsequently ingested for that time series, they will be returned as normal.
If no sample is found (by default) 5 minutes before a sampling timestamp, no value is returned for that time series at this point in time. This effectively means that time series "disappear" from graphs at times where their latest collected sample is older than 5 minutes or after they are marked stale.
Staleness will not be marked for time series that have timestamps included in their scrapes. Only the 5 minute threshold will be applied in that case.
If a query needs to operate on a very large amount of data, graphing it might time out or overload the server or browser. Thus, when constructing queries over unknown data, always start building the query in the tabular view of Prometheus's expression browser until the result set seems reasonable (hundreds, not thousands, of time series at most). Only when you have filtered or aggregated your data sufficiently, switch to graph mode. If the expression still takes too long to graph ad-hoc, pre-record it via a recording rule.
This is especially relevant for Prometheus's query language, where a bare
metric name selector like api_http_requests_total could expand to thousands
of time series with different labels. Also keep in mind that expressions which
aggregate over many time series will generate load on the server even if the
output is only a small number of time series. This is similar to how it would
be slow to sum all values of a column in a relational database, even if the
output value is only a single number.
This documentation is open-source. Please help improve it by filing issues or pull requests.