Update to collector main branch #1
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Collect system info and metrics for Azure Flexible Server and Cosmos DB for PostgreSQL. To start using this, the customer needs to supply a new config variable AZURE_SUBSCRIPTION_ID to the collector, as well as setting up the managed identity. When AZURE_SUBSCRIPTION_ID is not supplied, the collector doesn't error out and simply won't collect the system info.
…er with text format (#586) * Add auto_explain parameters support for JSON format * Support both new lines and Query Parameters with text format auto_explain better * Make sure not to include query params with explain JSON with normalized filter
In #494, I missed updating the --analyze-logfile flag. The flag is intended to analyze a logfile "offline", which does not work with the new mechanism, since that relies on being able to read the server's log_timezone and log_line_prefix. This adds additional flags to pass that information to the --analyze-logfile path. The timezone defaults to UTC; the prefix is required.
In passing, this also renames max_dead_tuples and num_dead_tuples to max_dead_item_ids/num_dead_item_ids, to track the upstream clarification on the naming.
Previously the collector would emit system catalogs that are only existing on EPAS, causing unnecessary information to show in the pganalyze dashboard. Additionally, EPAS allows function signatures to be re-used between functions and procedures (presumably this was allowed before upstream Postgres added CREATE PROCEDURE), which causes a server-side error with pganalyze. Since this is an edge case and there are no intents to support this upstream, ignore duplicate functions/procedures that have the same signature.
In some cases, it's possible for the user, database, or other free-form parts of the log line prefix to match too much of the input string, leaving the LogLevel to unintentionally match part of the content. E.g., with CloudSQL the vacuum analysis logs are output as > 2024-10-04 06:38:28.808 UTC [2569017]: [1-1] db=,user= LOG: automatic vacuum of table \"some_database.some_schema.some_table\": index scans: ... This is matched as user " LOG: automatic vacuum of table \"some_database.some_schema.some_table\": index" and log level "scans". The bogus log level is then mapped to 0, and the line is not handled correctly. To avoid this, change our regexp to only look for the log levels (and additional context strings that appear in place of log level on detail lines) we recognize.
RPM base packages have changed and no longer contain an init system by default. Since we mainly care about having a process running to then issue exec commands under, we can use a sleep here instead to avoid failure.
This asks for confirmation which we don't want in the automated flow.
* Bump protoc version to 28.2 * Update CI/release protoc version to 28.2 Also update the setup-protoc version to 3. Note that we still use the forked version to reduce a security risk: #308
…616) * Track cluster identifier (cluster ID) as part of system information This is intended to help identify servers that belong together, to allow the pganalyze app to group them in an overview, offer data in combined views, or allow Index Advisor to avoid considering indexes unused if they are used on a replica. To start with this identifies the cluster ID for given providers: - AWS: DBClusterIdentifier (user-assigned cluster name) - Azure: Source Server ID (replication parent server ID + resource group) - Crunchy Bridge: Cluster Parent ID (replication parent server ID) - Self hosted: pg_controldata system identifier (unique ID shared by all physical replicas) There are providers (e.g. GCP) and edge cases (e.g. when helper doesn't run for self hosted) where this doesn't work. Handling those cases, and allowing manual overrides of the cluster ID is left for a later follow-up.
Co-authored-by: Maciek Sakrejda <[email protected]>
The latter seems to no longer be available on Amazon Linux 2023. Since addgroup is just a convenience wrapper around groupadd and we are only using its basic functionality, we can switch to using groupadd directly.
Since snapshots include query references and query plan references, and since those can be orderered nondeterministically, the test can fail if the snapshot is built with fields in an unexpected order. The existing test tries to account for that by enumerating all possible combinations, but this is very difficult to maintain, and very difficult to scale to additional fields with this pattern. Instead, rewrite snapshots to a canonical format to ensure comparisons are stable. We can't just sort the array fields before serializing the snapshot for comparison, since QueryReferences and similar items are referenced by an index into their array. To work around that, we sort the QueryReferences (tracking their original index), and rewrite all entries that reference them to reference their new index instead. See discussion in #630 [1]. [1]: #630 (comment)
Co-authored-by: Keiko Oda <[email protected]>
* Make getFirstMetricValue to return the first value This function was previously returning the last value instead of the first value. This fix changes it and returns the value as soon as reading it. Co-authored-by: Maciek Sakrejda <[email protected]>
Bumps [golang.org/x/crypto](https://github.com/golang/crypto) from 0.25.0 to 0.31.0. - [Commits](golang/crypto@v0.25.0...v0.31.0) --- updated-dependencies: - dependency-name: golang.org/x/crypto dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
This adds the equivalent of the ini-style setting "max_buffer_cache_monitoring_gb" for the environment-based settings, which was missed when the setting was first added.
In passing, revise error text to include manual package install instructions
Fixes CVE-2024-45338, which is a DOS style vulnerability that's not a practical problem, but may be flagged by dependency scanners.
Some time ago (at least 9+ years) we introduced a maximum connection lifetime of 30 seconds via the SetConnMaxLifetime database/sql function. This was likely intended as an extra safety measure in case the collector didn't clean up its DB handle correctly. We've had a few minor issues with incorrect cleanups over the years (though usually not with the DB connection), but at this point our connection cleanup should be reliable. The connection lifetime causes an issue because it causes Go to reconnect to the database after 30 seconds have elapsed. This then effectively removes any connection local settings, in particular a statement_timeout. Effectively this meant that queries may unintentionally not have been subject to the statement_timeout, if the same connection previously took 30 seconds to do its work. To resolve, remove the max lifetime.
This executes on-demand query runs for the Query Tuning feature via a helper function that prevents reading the table data directly, as well as granting the collector user unnecessary permissions. This is now the only way that query runs can execute EXPLAIN ANALYZE, direct execution without the helper is intentionally not supported. In passing, expand the protobufs to pass down parameters and parameter types and optional planner settings, and refactor the query run code to handle work requiring a DB connection in a separate Go function. Further, this adds tests that utilize a new TEST_DATABASE_URL enviroment setting, which is used to run unit tests that require a database connection (previously only the integration tests used the database).
Like "--generate-stats-helper-sql" this allows generating a SQL script that installs the pganalyze.explain_analyze helper into all databases on a server, as specified by the server section name passed as an argument. Additionally, the "--generate-helper-explain-analyze-role" option allows setting the role that the helper function should be owned by as a value, defaulting to "pganalyze_explain". Note that the role creation, as well as any GRANT statements for the role must be taken care of separately.
#660) This acts as an additional defense in case the user does not revoke permissions from non-superuser for dblink as we recommend.
In some edge cases, for example unknown CONTEXT patterns, we may not detect the contents of a secondary line. If filtering of unidentified content is enabled, such lines need to be redacted.
This can occur in certain cases, for example auto_explain output on some Postgres versions. We previously would have not detected this CONTEXT line, but are now correctly detecting it as containing statement_parameter log secrets. In passing add a test case for the new "Query Parameters" field in auto_explain which gets redacted based on the statement_text filter that filters out the whole plan text (we normalize the individual fields in resulting query samples, but the log text for auto_explain is redacted altogether).
Because the buffercache logic initializes its own database (so it can run in parallel to other collection tasks), it needs to also ensure to close its DB handle so the connection is released again. In passing avoid the double assignment of "db" in the later parts of the function.
This makes it clear that we must clean up the database connection that was established, since we leak connections otherwise. To support this, refactor the log-based EXPLAIN code so defer can be used easily.
In almost all cases we have an existing database connection, so use that to retrieve the value of Postgres settings.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.