Select Git revision
Search by author
- Any Author
- authors
-
Arti Sandeep Dhareshwar adharesh
-
Benjamin Aaron Fleming
bflemin2
-
Brett Edward Caplan bcaplan2
-
Cedric William Prentice
cprentic
-
Chinhnam Dinh Nguyen
cn2149
-
Clara Madeleine Rigaux crigaux
-
Dale Morey dale
-
Daniel Wysling wysling
-
David Eisner
deisner
-
Daya Sagar dsagar
-
Devon Andrew Lai dlai12
-
Emily Li emilyli5
-
Eric Charles McKinney emckinne
-
Erica Lee Honadel ehonade1
-
Jason Alexander Steinberg
jsteinb4
-
Jim Zahniser zahniser
-
Karthik Valliappan Thiagarajan
kvthiaga
-
Kendrick Maurizio Montecino
kmtecino
-
Lynne Homann Cure lhcure
-
Megan Lok-Yi Nee mnee
This project is mirrored from https://github.com/metabase/metabase.
Pull mirroring updated .
- Aug 05, 2022
-
-
dpsutton authored
* Fix in-memory logger Our Admin > Troubleshooting > Logs page broke, just showing a spinner and never showing the logs. Don't quite understand why this fixes it. Javadocs are https://logging.apache.org/log4j/2.x/log4j-api/apidocs/org/apache/logging/log4j/LogManager.html#getContext-boolean- ```clojure logger=> (log/warn "test") nil logger=> (count @messages*) 0 ;; no in-memory logs so page is empty ;; change `(LogManager/getContext true)` in the momoized ns-logger fn ;; and then observe: logger=> (log/warn "test") nil logger=> (count @messages*) 4 ``` Some explorations that might shine some light: ```clojure logger=> (into {} (.getAppenders (.getLogger (LogManager/getContext false) (str *ns*)))) {} logger=> (into {} (.getAppenders (.getLogger (LogManager/getContext true) (str *ns*)))) {"metabase-appender" #object[metabase.logger.proxy$org.apache.logging.log4j.core.appender.AbstractAppender$ff19274a "0x4d680247" "metabase-appender"]} ``` So something is not hooked up quite right. * Add tests for metabase.logger * Ensure `api/util/logs` returns logs * tests * Check for presence of `MetabaseLoggerFactory` rather than whole str When in a namespace with a record, `Foo resolves to ns.Foo. But outside it resolves to ns/Foo. When running tests from the command line *ns* is user so it gets more complicated. * Kinda playing whackamole™ with `(LogManager/getContext true)` * Remove custom memoizing logger History: 39.2 we set `Multi-Release: true` in our manifest file and query speed drops like a stone. Jeff was able to track this to our logger calls in tight loops. We revert the multi-release and keep seeing the warning on startup > WARNING: sun.reflect.Reflection.getCallerClass is not supported. This will impact performance. Benchmarking on 39.2 (bench (log/trace "hi")) -> 15383 ns So we freaked out and set up a memoizing logger factory (bench (log/trace "hi")) -> 141 ns What a win. But we never noticed that the default *logger-factory* being picked up was slf4j ( `(log.impl/name log/*logger-factory*)` -> "org.slf4j" ). On 39.2 if you set the factory to the log4j2 version you get back to a great number: `(bench (log/trace "hi"))` -> 25 ns And thus ensuring that our logger is the default log4j2 version is even faster than our memoizing logger-factory. Memoizing factory: 141 ns slf4j factory: 2269 ns log4j2 factory: 31 ns What does `(LogManager/getContext false)` mean versus using `true`? We only need and want a single context. But log4j2 by default uses a context selector called `ClassLoaderContextSelector`. We could put all of this behind us if we used a context selector type `BasicContextSelector` but this is surprisingly hard to do: you have to set a system property. And since all of this stuff gets initialized in static initializers, we don't have an opportunity to do this programmatically. The way around this is to require people to pass this system property on startup which is not acceptable. So getContext true checks for a threadlocal context in a specific static variable and falls back to a Default context. getContext false looks at classloaders and ends up at a different context. BUT: the log.impl version uses a closure over getContext false instead of getting it each time. And I suspect that when it does this there is only one so it is the default and continues to use this one. In our LoggerFactory implementation we were looking up the context each time. This still seems to work and everything is playing nice in our application classloader but its totally possible that our drivers are not hitting this. I'll have to investigate this. Verification: - build the uberjar locally (`bin/build`) - copy to some temp directory and also copy criterium.jar ```shell MB_JETTY_PORT=4000 java "$(socket-repl 4001)" -cp locally-built.jar:criterium.jar metabase.core ``` ```clojure /tmp/locally-built via
-
- Jul 28, 2022
-
-
dpsutton authored
* Try multi-release true again in our manifest Problem statement: Luiz packs our partner jars (exasol, starburst, etc.) into our jar so they can be "first class" and in cloud. But with the 44 cycle we've run into some issues: ```shell /tmp/j via
Cam Saul <1455846+camsaul@users.noreply.github.com>
-
- Apr 19, 2022
-
-
Braden Shepherdson authored
* Make namespace aliasing consistent everywhere; enforce with clj-kondo See the table of aliases in .clj-kondo/config.edn Notable patterns: - `[metabase.api.foo :as api.foo]` - `[metabase.models.foo :as foo]` - `[metabase.query-processor.foo :as qp.foo]` - `[metabase.server.middleware.foo :as mw.foo]` - `[metabase.util.foo :as u.foo]` - `[clj-http.client :as http]` and `[metabase.http-client :as client]` Fixes #19930.
-
- Mar 14, 2022
-
-
Cam Saul authored
Upgrade Liquibase to latest version; remove final Java source file and need for `clojure -X:deps prep` (#20611) * Upgrade Liquibase to latest version * Try adjusting log * Fix checksums for the TWO migrations with ID = 32 * FINALLY get Liquibase to use Log4j2 * Set Liquibase ConsoleUIService OutputStream to null OutputStream * Manually define a package for our H2 proxy class so Java 8 works * Fix package-name determination code * Update migrations file spec * `databasechangelog` shouldn't be upper-case * Lower-case quartz table names * More MySQL fixes
-
- Jul 30, 2021
-
-
Cam Saul authored
-
