PostgreSQL Weekly News - October 10, 2021

Posted on 2021-10-11 by PWN

PostgreSQL Weekly News - October 10, 2021

PostgreSQL Product News

pgCluu 3.2, a Perl program to audit PostgreSQL performance, released.

PGroonga 2.3.2 a full text search platform for all languages, released.

PostgreSQL Jobs for October

PostgreSQL in the News

Planet PostgreSQL:

PostgreSQL Weekly News is brought to you this week by David Fetter

Submit news and announcements by Sunday at 3:00pm PST8PDT to

Applied Patches

Michaël Paquier pushed:

  • Fix snapshot builds during promotion of hot standby node with 2PC. Some specific logic is done at the end of recovery when involving 2PC transactions: 1) Call RecoverPreparedTransactions(), to recover the state of 2PC transactions into memory (re-acquire locks, etc.). 2) ShutdownRecoveryTransactionEnvironment(), to move back to normal operations, mainly cleaning up recovery locks and KnownAssignedXids (including any 2PC transaction tracked previously). 3) Switch XLogCtl->SharedRecoveryState to RECOVERY_STATE_DONE, which is the tipping point for any process calling RecoveryInProgress() to check if the cluster is still in recovery or not. Any snapshot taken between steps 2) and 3) would be empty, causing any transaction relying on a snapshot at this point to potentially corrupt data as there could still be some 2PC transactions to track, with RecentXmin moving backwards on successive calls to GetSnapshotData() in the same transaction. As SharedRecoveryState is the point to take into account to know if it is safe to discard KnownAssignedXids, this commit moves step 2) after step 3), so as we can never finish with empty snapshots. This exists since the introduction of hot standby, so backpatch all the way down. The window with incorrect snapshots is extremely small, but I have seen it when running, as did buildfarm member fairywren. Thomas Munro also found it independently. Special thanks to Andres Freund for taking the time to analyze this issue. Reported-by: Thomas Munro, Michael Paquier Analyzed-by: Andres Freund Discussion: Backpatch-through: 9.6

  • Fix warning in TAP test of pg_verifybackup. Oversight in a3fcbcd. Reported-by: Thomas Munro Discussion: Backpatch-through: 13

  • Refactor per-destination file rotation in logging collector. stderr and csvlog have been using duplicated code when it came to the rotation of their file by size, age or if forced by a user request (pg_ctl logrotate or the SQL function pg_rotate_logfile). The main difference between both is that stderr requires its file to always be opened, so as it is possible to have a redirection route if the logging collector is not ready yet to do its work if alternate destinations are enabled. Also, if csvlog gets disabled, we need to close properly its meta-data stored in the logging collector (last file name for current_logfiles and fd currently open for business). Except for those points, the code is the same in terms of error handling and if a file should be created or just continued. This change makes the code simpler overall, and it will help in the introduction of more file-based log destinations. This refactoring is similar to the work done in 5b0b699. Most of the duplication originates from fd801f4. Some of the TAP tests of pg_ctl check the case of a forced log rotation, but this is somewhat limited as there is no coverage for log_rotation_age or log_rotation_size (these may not be worth the extra resources to run either), and no coverage for reload of log_destination with different combinations of stderr and csvlog. I have tested all those cases separately for this refactoring. Author: Michael Paquier Discussion:

  • Fix compilation warning in syslogger.c. Oversight in 5c6e33f. Author: Nathan Bossart Discussion:

  • Refactor fallback to stderr for csvlog to handle better WIN32 service case. send_message_to_server_log() would force a redirection of a log entry to stderr in some cases for csvlog, like the syslogger not being available yet. If this happens, csvlog would fall back to stderr to log some information rather than nothing. The code was organized so as stderr is done before csvlog, with csvlog checking that stderr did not happen yet with a reversed condition. With this code organization, it could be possible to lose some messages if running Postgres as a service on WIN32, as there is no usable stderr, and the handling of the StringInfoData holding the message for stderr was rather confusing because of that. This commit moves the csvlog handling to be before stderr, as as we are able to track down if it is necessary to log something to stderr. The reduces the handling of stderr to be in a single code path, adding a fallback to event logs for a WIN32 service. This also simplifies the way we handle the StringInfoData for stderr, making easier the integration of new file-based log destinations. I got to play with services and event logs on Windows while checking this change. Reviewed-by: Chris Bandy Discussion:

Daniel Gustafsson pushed:

Peter Eisentraut pushed:

Tom Lane pushed:

Andres Freund pushed:

Bruce Momjian pushed:

Fujii Masao pushed:

Amit Kapila pushed:

Robert Haas pushed:

Dean Rasheed pushed:

  • Fix corner-case loss of precision in numeric_power(). This fixes a loss of precision that occurs when the first input is very close to 1, so that its logarithm is very small. Formerly, during the initial low-precision calculation to estimate the result weight, the logarithm was computed to a local rscale that was capped to NUMERIC_MAX_DISPLAY_SCALE (1000). However, the base may be as close as 1e-16383 to 1, hence its logarithm may be as small as 1e-16383, and so the local rscale needs to be allowed to exceed 16383, otherwise all precision is lost, leading to a poor choice of rscale for the full-precision calculation. Fix this by removing the cap on the local rscale during the initial low-precision calculation, as we already do in the full-precision calculation. This doesn't change the fact that the initial calculation is a low-precision approximation, computing the logarithm to around 8 significant digits, which is very fast, especially when the base is very close to 1. Patch by me, reviewed by Alvaro Herrera. Discussion:

Etsuro Fujita pushed: