* Add OpenAPI 3.2 specification generation for Prometheus HTTP API
This commit introduces an OpenAPI specification for the Prometheus API.
After testing multiple code-generation servers with built-in APIs, this
implementation uses an independent spec file outside of the critical path.
This spec file is tested with a framework present in this pull request.
The specification helps clients know which parameters they can use and is
served at /api/v1/openapi.yaml. The spec file will evolve with the
Prometheus API and has the same version number.
Downstream projects can tune the APIs presented in the spec file with
configuration options using the IncludePaths setting for path filtering.
In the future, there is room to generate a server from this spec file
(e.g. with interfaces), but this is out of scope for this pull request.
Architecture:
- Core OpenAPI infrastructure (openapi.go): Dynamic spec building,
caching, and thread-safe spec generation
- Schema definitions (openapi_schemas.go): Complete type definitions
for all API request and response types
- Path specifications (openapi_paths.go): Endpoint definitions with
parameters, request bodies, and response schemas
- Examples (openapi_examples.go): Realistic request/response examples
- Helper functions (openapi_helpers.go): Reusable builders for common
OpenAPI structures
Testing:
- Comprehensive test suite with golden file validation
- Test helpers package for API testing infrastructure
- OpenAPI compliance validation utilities
The golden file captures the complete specification for snapshot testing.
Update with: go test -run TestOpenAPIGolden -update-openapi-spec
REVIEWERS: The most important thing to check would be the OpenAPI golden
file (web/api/v1/testdata/openapi_golden.yaml). Test scenarios are important
as they test the actual OpenAPI spec validity.
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
* Add OpenAPI 3.1 support with version selection
Add support for both OpenAPI 3.1 and 3.2 specifications with version
selection via openapi_version query parameter. Defaults to 3.1 for
broader compatibility
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
* Enhance OpenAPI examples and add helper functions
- Add timestampExamples helper for consistent time formatting
- Add exampleMap helper to simplify example creation
- Improve example summaries with query details
- Add matrix result example for range vector queries
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
* web/api: Add AtST method to test helper iterators
Implement the AtST() method required by chunkenc.Iterator interface
for FakeSeriesIterator and FakeHistogramSeriesIterator test helpers.
The method returns 0 as these test helpers don't use start timestamps
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
* OpenAPI: Add minimum coverage test
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
* OpenAPI: Improve examples handling
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
---------
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
* UI: Fix broken Y axis after graph page reload
The new `y_axis_min` setting was always encoded into the URL, even if its value
was `null` (in which case it would be set to an empty string parameter). On the
decoding side, this wasn't taken into account correctly, and we tried to parse
the empty string as a float, causing completely broken graphs showing nothing
after reloading the graph page with such URL parameters.
I'm doing two things now:
* For the future, only encode the Y axis min into the URL if it's set at all,
similar as we do for the `end_input` and `moment_input` fields.
* On the decoding side, accommodate people (at least for now) who already saved
some links with the empty `y_axis_min` parameter by treating an empty string
as `null` instead of a number.
Signed-off-by: Julius Volz <julius.volz@gmail.com>
* Add URL state encoding/decoding tests
Signed-off-by: Julius Volz <julius.volz@gmail.com>
---------
Signed-off-by: Julius Volz <julius.volz@gmail.com>
When accepting an autocompletion result within an Identifier node (could be a
metric name, function name, keyword, etc.), the inserted completion should
replace the entire Identifier node all the way to its last character, not only
to the current cursor position.
A limitation is that the correct replacement-until-end-of-identifier only works
when e.g. a function name is currently incomplete (which is likely anyway when
trying to replace it with a different one). This is because otherwise the
Identifier node gets replaced with a more specific function node type (like
`Rate`, `SumOverTime`, etc.), and handling all those adds more complexity.
https://github.com/prometheus/prometheus/issues/15839
Signed-off-by: Julius Volz <julius.volz@gmail.com>
The /classic/static/* route was added to serve vendor JavaScript and CSS
files (jQuery, Bootstrap, etc.) for console templates. These vendor assets
were removed in #14807 due to security vulnerabilities, making this route
obsolete as it now serves an empty directory.
The console feature remains functional via --web.console.templates and
--web.console.libraries flags. Users who need JavaScript/CSS libraries in
their custom console templates must provide these assets within the
directory specified by --web.console.libraries.
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
When React components mount/unmount repeatedly, each creating a new
PromQLExtension, memory leaks occur due to LRU caches with ttlAutopurge
timers keeping references alive and in-flight HTTP requests holding
closure references. This adds destroy() methods throughout the class
hierarchy to properly release resources on unmount.
Changes:
- Add destroy() to PrometheusClient interface (optional)
- HTTPPrometheusClient: track AbortControllers and abort pending requests
- Cache: clear all LRU caches and reset cached data
- CachedPrometheusClient: delegate to cache and underlying client
- HybridComplete: delegate to prometheusClient
- CompleteStrategy: add optional destroy() method
- PromQLExtension: delegate to complete strategy
Signed-off-by: Ben Blackmore <ben.blackmore@dash0.com>
Update generated files with latest functions from Prometheus, adding
support for first_over_time, info, ts_of_first_over_time,
ts_of_last_over_time, ts_of_max_over_time, and ts_of_min_over_time.
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
Add make targets to generate and check PromQL function signatures and
documentation for the Mantine UI. The generate-promql-functions target
runs the Go generators and automatically lints the output files. The
check-generated-promql-functions target verifies that generated files
are up to date, similar to check-generated-parser.
Fix the gen_functions_list generator to output properly formatted
TypeScript code with correct indentation and semicolons.
Add check-generated-promql-functions to the UI tests CI job to ensure
generated files stay in sync with upstream changes.
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
By running `make update-all-go-deps`.
`hashicorp/consul/api` must be held at v1.32.1 because later versions
require Go 1.25 and we choose to ensure that Promethes builds with the
last two versions of Go.
Also: fix compilation errors in remote-write example.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>
Add a maximum limit of 10,000 to the TSDB status endpoint to prevent
resource exhaustion from excessively large limit values, as we preallocate
[]Stat for up to the limit: `make([]Stat, 0, length)`.
Note that the endpoint acquires a cardinality mutex during
stats calculation, so this can not be run in parallel.
Signed-off-by: Julien Pivotto <291750+roidelapluie@users.noreply.github.com>
Currently both the backend and frontend printers/formatters/serializers
incorrectly transform the following expression:
```
up * ignoring() group_left(__name__) node_boot_time_seconds
```
...into:
```
up * node_boot_time_seconds
```
...which yields a different result (including the metric name in the result
vs. no metric name).
We need to keep empty `ignoring()` modifiers if there is a grouping modifier
present.
Signed-off-by: Julius Volz <julius.volz@gmail.com>
