Fixing a couple bugs, adds some tests and hopefully decrease
lock contention when logging.
Switching from using sync.RWMutex to go-kit SwapLogger.
Fixes bug when creating a new logger from an existing one that
screwed up the keyvals and/or lost the logger name.
Ref #44681
Moves the request tracing middleware earlier in the chain, just after the tracing middleware
and before the log middleware. With these changes we'll be able to track
authentication/authorization status failures that currently exits early and don't execute the
request tracing middleware. In addition, there might be some other routes now being tracked
with this that we didn't do before.
Fixes#39590
Fixes so that errors (directory not exists, no permission) when scanning plugins are logged as
errors rather than with debug level. In addition, before the scanning would halt in case of
referenced errors, but with these changes the scanning will continue. If any other error
than the referenced error happens the scanning for specific directory would halt and return
the error, e.g. stop Grafana from starting.
Fixes#43012
Moves/refactor Grafana specific functionality related to plugin dashboards
out to specific services for importing dashboards and keep app plugin dashboards
up-to-date.
Fixes#44257
Fix duplicate validation when multiple organizations have been configured.
This makes sure that all duplicate validation is being made for each organization.
Fixes#44126
Fixes so that searching for folder id zero in folder/dashboard search returns
dashboards located in the general folder and not including all folders as it
did before.
Fixes#40273
Get rid of using legacydata.RequestHandler in HTTPServer, /api/tsdb/query and pkg/expr
with the goal of deprecating /api/tsdb/query and remove it completely eventually. This is
the first step of cleaning up the HTTP API query endpoint.
Co-authored-by: Marcus Efraimsson <marcus.efraimsson@gmail.com>
Co-authored-by: Alexander Emelin <frvzmb@gmail.com>
Split the nightly Docker scan job into separate pipelines, one for each
image, to easier understand which images fails/includes high/critical
vulnerabilities.
Refactor usage of legacy data contracts. Moves legacy data contracts
to pkg/tsdb/legacydata package.
Refactor pkg/expr to be a proper service/dependency that can be provided
to wire to remove some unneeded dependencies to SSE in ngalert and other places.
Refactor pkg/expr to not use the legacydata,RequestHandler and use
backend.QueryDataHandler instead.
When a request going through Grafana data source proxy responds with a websocket
upgrade response we cannot override the response body since it produces an error.
This problem seems to have been introduced in Grafana v8.0 by #38962.
In addition #40303 added same problem.
Fixes#41292
Improves error handling when installing plugins by checking for
error before adding a defer of closing of the zip reader to not
create a panic when there's an invalid zip file.
Fixes#41029
Clarifies documentation in regards to data source provisioning and certain
settings that earlier was explained as applicable to All data sources.
Fixes#19574
Co-authored-by: Will Browne <wbrowne@users.noreply.github.com>
Require guardian.New to take context.Context as first argument.
Migrates the GetDashboardAclInfoListQuery to be dispatched using context.
Ref #36734
Co-authored-by: Emil Tullstedt <emil.tullstedt@grafana.com>
Co-authored-by: sam boyer <sam.boyer@grafana.com>
Fix bug with engine cache not updating after data source updated, introduced in #36635.
Removes the SQL engine cache because since #36635 we already have instance caching in place using the SDK instance manager.
This also adds support for disposing cached SQL/xorm engines before a query happens to a data source that has been updated. Done by letting the SQL engine implementing the InstanceDisposer interface.
Fixes#37299
* Chore: GetDashboardQuery should be dispatched using DispatchCtx
* Fix after merge
* Changes after review
* Various fixes
* Use GetDashboardCtx function instead of GetDashboard
Adds a new setting dataproxy.row_limit that allows an operator to limit the
amount of rows being processed/accepted in response to database queries
originating from SQL data sources.
Closes#38975
Ref #39095
Co-authored-by: achatterjee-grafana <70489351+achatterjee-grafana@users.noreply.github.com>
Fix region annotations not displayed correctly when returning timeend column
as epoch timestamp and by that making sure that the returned data frame field
named timeend is treated as time type.
Fixes#38533
Skipping TestDevenvDashboardValidity test for now to resolve broken
test-backend step in main/branches. Did also run scripts/stripnulls.sh
hence the devenv dashboard changes.
If DispatchCtx is called and there's a handler registered using
AddHandler we now allow that and not return a handler not
found error. In addition we log a warning message helping
the developers know that AddHandlerCtx should be used
instead.
If Dispatch is called and there's a handler registered
using AddHandlerCtx we log a warning message helping
the developers know that DispatchCtx should be used instead.
Make sure that default timeout settings are based on configuration
parameters. This now applies for core data sources using old TSDB
contracts and new SDK contracts. Before it was only applied for old TSDB
contracts.
Also moves global setting variables to non-global (setting.Cfg).
Name of time field changed in v8 for time series queries from Time to the name of the selected
time column, i.e. time or time_sec. These changes should make sure that the name of time field
is always returned as Time for time series queries.
Fixes#36059
Co-authored-by: Ryan McKinley <ryantxu@gmail.com>
Fixes panic/fatal error concurrent map writes in SQL data sources when multiple
queries are executed concurrently and you interpolate SQL query before executing it.
Fixes#35469
#31871 introduced support for configuring timeout in seconds
for HTTP data sources. That had a bug where backend expected
a numeric timeout value where it was actually stored as a
string. This should resolve this by requiring input to be
numbers, storing input as numeric and falling back to string
value if there's no numeric value.
Ref #31871
Fixes a problem with query variables where SQL query returning for example only
numeric values didn't populate the query variables with values.
Fixes#35391
Move QueryData method into backend plugin manager which HandleRequest uses to
query data from plugin SDK supported data sources. This allowed us to remove a lot
of code no longer needed.
Ref #21510
Co-authored-by: Will Browne <wbrowne@users.noreply.github.com>
Uses new httpclient package from grafana-plugin-sdk-go introduced
via grafana/grafana-plugin-sdk-go#328.
Replaces the GetHTTPClient, GetTransport, GetTLSConfig methods defined
on DataSource model.
Longer-term the goal is to migrate core HTTP backend data sources to use the
SDK contracts and using httpclient.Provider for creating HTTP clients and such.
Co-authored-by: Arve Knudsen <arve.knudsen@gmail.com>
In the case permissions has been added on dashboard(s). Later permissions for the
parent folder of the dashboard is edited in such a way that dashboard in that folder
has a permission that is a duplicate of an inherited one. This PR changes so that
duplicate permissions are now filtered out from /api/dashboards/id/<dashboard id>/permissions.
Duplicate permission are not filtered out if the permission on dashboard is higher
than on the inherited folder.
Fixes#33296
Co-authored-by: Arve Knudsen <arve.knudsen@gmail.com>
Before these changes the request tracing was added for each route
registered using the routing.RouteRegister, see code. This had the
consequence that middleware executed earlier/later in the request
pipeline was not part of the request tracing middleware life-cycle
which measures the duration of requests among other things.
In the logger middleware we do extract the current distributed trace
identifier, if available, and set that on request info/error log messages.
With these changes we can extract the current distributed trace identifier,
if available, and set that on the contextual HTTP request logger
(models.ReqContext.Logger) which would improve the possibility to correlate
all HTTP request log messages with traces.
In addition, the request tracing middleware is now executed first and last in
the request pipeline and should therefore result in more accurate timing
measurements (request duration).
Co-authored-by: Arve Knudsen <arve.knudsen@gmail.com>
For security reasons, log any SQL connection error
details rather than returning it to the client.
Fixes#26623Fixes#22000
Co-authored-by: Arve Knudsen <arve.knudsen@gmail.com>
Fix encoded characters in URL path should be proxied as encoded in the data proxy.
Fixes#26870Fixes#31438
Co-authored-by: Arve Knudsen <arve.knudsen@gmail.com>
Marcus Efraimsson
