nick-boss-tech commented on code in PR #5348: URL: https://github.com/apache/solr/pull/5348#discussion_r4637774336
########## solr/bin/solr.cmd: ########## @@ -1292,6 +1304,12 @@ set eout=%eout:"=% echo !eout! GOTO :eof +:delete_node_files + REM Remove the files recording the node on the given port + del /q "%SOLR_PID_DIR%\solr-%1.pid" >nul 2>nul Review Comment: 🤖 *AI text below* 🤖 Could we exclude directories from PID enumeration and make this helper refuse directory targets for both marker paths? `dir /b` can return directories, and `stop --all` invokes cleanup even if the marker couldn’t be read. On Windows, [`del /q` given a directory deletes the files inside it](https://learn.microsoft.com/en-us/windows-server/administration/windows-commands/del). A directory with a PID-shaped name could therefore lose its contents even though no Solr process was stopped. The helper guard also protects the direct `stop -p` path. ########## solr/bin/solr.cmd: ########## @@ -1200,6 +1201,17 @@ IF "%FG%"=="1" ( set "SCRIPT_ERROR=Solr did not start or was not reachable. Check the logs for errors." goto err ) + + REM START /B leaves us without a PID, so ask netstat who owns the port now that Solr is up Review Comment: 🤖 *AI text below* 🤖 Could we tie PID publication to the JVM this invocation launched rather than to bare `status --max-wait-secs`? That command takes a snapshot of all local Solr processes: an empty snapshot can succeed before the new JVM appears, and an unrelated unhealthy node can make the command fail before the PID is written. Even a timeout for this node can leave a JVM that comes up later. The old `.port` record was written before the wait; the new `stop --all` ignores it. Please scope readiness to the target node and keep a recovery path for a verified child that outlives a failed or interrupted start. ########## solr/bin/solr.cmd: ########## @@ -875,7 +876,7 @@ IF "%SOLR_PORT_LISTEN%"=="" ( @echo Stopping Solr process %%N running on port %SOLR_PORT_LISTEN% IF "%STOP_PORT%"=="" set /A STOP_PORT=%SOLR_PORT_LISTEN% - 1000 "%JAVA%" %SOLR_SSL_OPTS% %SOLR_TOOL_OPTS% -Djetty.home="%SOLR_SERVER_DIR%" -jar "%SOLR_SERVER_DIR%\start.jar" %SOLR_JETTY_CONFIG% STOP.PORT=!STOP_PORT! STOP.KEY=%STOP_KEY% --stop - del "%SOLR_PID_DIR%"\solr-%SOLR_PORT_LISTEN%.port + CALL :delete_node_files %SOLR_PORT_LISTEN% Review Comment: 🤖 *AI text below* 🤖 Could we move this cleanup after a confirmed exit and return a failure if the process is still alive? This removes the `.pid` immediately after sending the graceful stop request. If that request fails and the fallback cannot terminate the process (for example, if `jstack` fails, the `&&` chain never reaches `taskkill`), the script still reaches its normal success exit. This carries the old eager `.port` cleanup over to the new authoritative `.pid`: a subsequent `stop --all` has no record to recover with. Please keep the tracking record until the node is gone. ########## solr/core/src/java/org/apache/solr/cli/SolrProcessManager.java: ########## @@ -120,16 +119,8 @@ public Collection<SolrProcess> scanSolrPidFiles() throws IOException { Files.list(pidDir) .filter(p -> pidFilePattern.matcher(p.getFileName().toString()).matches())) { for (Path p : pidFiles.toList()) { - Optional<SolrProcess> process; - if (p.toString().endsWith(".port")) { - // On Windows, the file is a 'PORT' file containing the port number. - Integer port = Integer.valueOf(Files.readAllLines(p).getFirst()); - process = processForPort(port); - } else { - // On Linux, the file is a 'PID' file containing the process ID. - Long pid = Long.valueOf(Files.readAllLines(p).getFirst()); - process = getProcessForPid(pid); - } + Long pid = Long.valueOf(Files.readAllLines(p).getFirst().trim()); Review Comment: 🤖 *AI text below* 🤖 Could the Windows writer publish the PID file with a temporary file plus a rename, and could this scan skip empty or malformed records? The new `>` redirect creates or truncates the final `.pid` before writing its contents. A concurrent `solr status` can see an empty file, throw at `getFirst()`, and never list the other healthy nodes. A test with an empty `.pid` next to a valid one would cover the reader side. ########## solr/bin/solr.cmd: ########## @@ -830,35 +830,36 @@ IF "%SOLR_PORT_LISTEN%"=="" ( IF "%STOP_ALL%"=="1" ( REM Stop all running Solr instances set found_it=0 - for /f "usebackq" %%i in (`dir /b "%SOLR_PID_DIR%" ^| findstr /i "^solr-.*\.port$"`) do ( - set SOME_SOLR_PORT= - For /F "delims=" %%J In ('type "%SOLR_PID_DIR%\%%i"') do set SOME_SOLR_PORT=%%~J - if NOT "!SOME_SOLR_PORT!"=="" ( - for /f "tokens=2,5" %%j in ('netstat -aon ^| find "TCP " ^| find ":0 " ^| find ":!SOME_SOLR_PORT! "') do ( - @REM j is the ip:port and k is the pid - IF NOT "%%k"=="0" ( - IF "%%j"=="%SOLR_HOST_BIND%:!SOME_SOLR_PORT!" ( - set found_it=1 - @echo Stopping Solr process %%k running on port !SOME_SOLR_PORT! - IF "%STOP_PORT%"=="" ( - set /A LOCAL_STOP_PORT=!SOME_SOLR_PORT! - 1000 - ) else ( - set LOCAL_STOP_PORT=%STOP_PORT% - ) - "%JAVA%" %SOLR_SSL_OPTS% -Djetty.home="%SOLR_SERVER_DIR%" -jar "%SOLR_SERVER_DIR%\start.jar" STOP.PORT=!LOCAL_STOP_PORT! STOP.KEY=%STOP_KEY% --stop - del "%SOLR_PID_DIR%"\solr-!SOME_SOLR_PORT!.port - REM wait for the process to terminate - CALL :wait_for_process_exit %%k !SOLR_STOP_WAIT! - REM Kill it if it is still running after the graceful shutdown - IF EXIST "%JAVA_HOME%\bin\jstack.exe" ( - qprocess "%%k" >nul 2>nul && "%JAVA_HOME%\bin\jstack.exe" %%k && taskkill /f /PID %%k - ) else ( - qprocess "%%k" >nul 2>nul && taskkill /f /PID %%k - ) - ) - ) + for /f "usebackq" %%i in (`dir /b "%SOLR_PID_DIR%" ^| findstr /i "^solr-[0-9][0-9]*\.pid$"`) do ( Review Comment: 🤖 *AI text below* 🤖 Could we preserve `stop --all` for foreground Windows starts? The old foreground branch wrote a `.port` file that this loop could use. The new branch writes neither file, and this loop only looks for `.pid`, so an otherwise healthy foreground node started from another console is skipped. If this is an intentional behavior change, please document `stop -p` as the alternative and add a Windows lifecycle test. ########## solr/solr-ref-guide/modules/upgrade-notes/pages/major-changes-in-solr-10.adoc: ########## @@ -91,6 +91,11 @@ A request with no other parameters returns every collection. The v1 `CLUSTERSTATUS` command is unchanged, and `includeAll` still defaults to true. +=== Windows PID Files + +On Windows, `bin\solr.cmd` now writes a true `solr-<port>.pid` file instead of relying on the `solr-<port>.port` file. +The `solr-<port>.port` file is deprecated and will be removed in a future version. Review Comment: 🤖 *AI text below* 🤖 Could the upgrade notes distinguish deprecating writes from dropping reads immediately? A Windows node started with the old script has only a `.port` file, and `stop --all` in the new script won’t find it. If there’s no migration fallback, please tell operators to stop those nodes with the old script before switching, or use an explicit `stop -p` and then start them with the new script to create the `.pid` file. ########## solr/core/src/java/org/apache/solr/cli/SolrProcessManager.java: ########## @@ -120,16 +119,8 @@ public Collection<SolrProcess> scanSolrPidFiles() throws IOException { Files.list(pidDir) .filter(p -> pidFilePattern.matcher(p.getFileName().toString()).matches())) { for (Path p : pidFiles.toList()) { - Optional<SolrProcess> process; - if (p.toString().endsWith(".port")) { - // On Windows, the file is a 'PORT' file containing the port number. - Integer port = Integer.valueOf(Files.readAllLines(p).getFirst()); - process = processForPort(port); - } else { - // On Linux, the file is a 'PID' file containing the process ID. - Long pid = Long.valueOf(Files.readAllLines(p).getFirst()); - process = getProcessForPid(pid); - } + Long pid = Long.valueOf(Files.readAllLines(p).getFirst().trim()); + Optional<SolrProcess> process = getProcessForPid(pid); Review Comment: 🤖 *AI text below* 🤖 Could we revalidate liveness before deleting a file that isn’t in `pidProcessMap`? The map is built before `scanSolrPidFiles()` lists the directory. A concurrent Windows start can launch and atomically publish a correct PID file between those two steps. This scan then sees the new record, treats the snapshot miss as proof the process is gone, and deletes the file for a live node that `stop --all` now depends on. Atomic publication alone doesn’t address this; before deleting, also make sure the record hasn’t been replaced since it was read. A test that creates the manager before starting and registering a mock process would cover it. ########## solr/bin/solr.cmd: ########## @@ -830,35 +830,36 @@ IF "%SOLR_PORT_LISTEN%"=="" ( IF "%STOP_ALL%"=="1" ( REM Stop all running Solr instances set found_it=0 - for /f "usebackq" %%i in (`dir /b "%SOLR_PID_DIR%" ^| findstr /i "^solr-.*\.port$"`) do ( - set SOME_SOLR_PORT= - For /F "delims=" %%J In ('type "%SOLR_PID_DIR%\%%i"') do set SOME_SOLR_PORT=%%~J - if NOT "!SOME_SOLR_PORT!"=="" ( - for /f "tokens=2,5" %%j in ('netstat -aon ^| find "TCP " ^| find ":0 " ^| find ":!SOME_SOLR_PORT! "') do ( - @REM j is the ip:port and k is the pid - IF NOT "%%k"=="0" ( - IF "%%j"=="%SOLR_HOST_BIND%:!SOME_SOLR_PORT!" ( - set found_it=1 - @echo Stopping Solr process %%k running on port !SOME_SOLR_PORT! - IF "%STOP_PORT%"=="" ( - set /A LOCAL_STOP_PORT=!SOME_SOLR_PORT! - 1000 - ) else ( - set LOCAL_STOP_PORT=%STOP_PORT% - ) - "%JAVA%" %SOLR_SSL_OPTS% -Djetty.home="%SOLR_SERVER_DIR%" -jar "%SOLR_SERVER_DIR%\start.jar" STOP.PORT=!LOCAL_STOP_PORT! STOP.KEY=%STOP_KEY% --stop - del "%SOLR_PID_DIR%"\solr-!SOME_SOLR_PORT!.port - REM wait for the process to terminate - CALL :wait_for_process_exit %%k !SOLR_STOP_WAIT! - REM Kill it if it is still running after the graceful shutdown - IF EXIST "%JAVA_HOME%\bin\jstack.exe" ( - qprocess "%%k" >nul 2>nul && "%JAVA_HOME%\bin\jstack.exe" %%k && taskkill /f /PID %%k - ) else ( - qprocess "%%k" >nul 2>nul && taskkill /f /PID %%k - ) - ) - ) + for /f "usebackq" %%i in (`dir /b "%SOLR_PID_DIR%" ^| findstr /i "^solr-[0-9][0-9]*\.pid$"`) do ( + set "SOME_SOLR_PORT=%%~ni" + set "SOME_SOLR_PORT=!SOME_SOLR_PORT:solr-=!" + set SOME_SOLR_PID= + For /F "delims=" %%J In ('type "%SOLR_PID_DIR%\%%i"') do set SOME_SOLR_PID=%%~J + REM Only stop the PID from the file if it still listens on the port the file is named after + set PID_OWNS_PORT=0 + for /f "tokens=2,5" %%j in ('netstat -aon ^| find "TCP " ^| find ":0 " ^| find ":!SOME_SOLR_PORT! "') do ( + @REM j is the ip:port and k is the pid + IF "%%j"=="%SOLR_HOST_BIND%:!SOME_SOLR_PORT!" IF "%%k"=="!SOME_SOLR_PID!" set PID_OWNS_PORT=1 + ) + if "!PID_OWNS_PORT!"=="1" ( + set found_it=1 + @echo Stopping Solr process !SOME_SOLR_PID! running on port !SOME_SOLR_PORT! + IF "%STOP_PORT%"=="" ( + set /A LOCAL_STOP_PORT=!SOME_SOLR_PORT! - 1000 + ) else ( + set LOCAL_STOP_PORT=%STOP_PORT% + ) + "%JAVA%" %SOLR_SSL_OPTS% -Djetty.home="%SOLR_SERVER_DIR%" -jar "%SOLR_SERVER_DIR%\start.jar" STOP.PORT=!LOCAL_STOP_PORT! STOP.KEY=%STOP_KEY% --stop + REM wait for the process to terminate + CALL :wait_for_process_exit !SOME_SOLR_PID! !SOLR_STOP_WAIT! + REM Kill it if it is still running after the graceful shutdown + IF EXIST "%JAVA_HOME%\bin\jstack.exe" ( + qprocess "!SOME_SOLR_PID!" >nul 2>nul && "%JAVA_HOME%\bin\jstack.exe" !SOME_SOLR_PID! && taskkill /f /PID !SOME_SOLR_PID! + ) else ( + qprocess "!SOME_SOLR_PID!" >nul 2>nul && taskkill /f /PID !SOME_SOLR_PID! ) ) + CALL :delete_node_files !SOME_SOLR_PORT! Review Comment: 🤖 *AI text below* 🤖 Could we remove the files only after confirming the recorded process stopped or the record is stale? This call also runs when `PID_OWNS_PORT` is `0`, which can simply mean that this invocation used a different `SOLR_HOST_BIND` from the one used at startup. The previous loop left the `.port` file alone on a bind mismatch; now a stop with the wrong include file removes the record a correctly configured retry needs. Also tie cleanup to the specific PID record inspected: if a replacement starts during the shutdown wait and publishes a new file on the same port, this unconditional delete removes the replacement’s record. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
