model-writtenrun 6091 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Apr 2021 · 21 entries
- Jacob Champion proposed retaining each user's original authenticated identity before role mapping, initially exposing it through log_line_prefix so administrators could audit multiple external users sharing one database role #1.
- Stephen Frost, Tom Lane, and Magnus Hagander questioned repeated logging cost and identity visibility; discussion converged on logging the identity once through log_connections, while preserving the original identity across role renames #2 #3 #5 #8 #11.
- The discussion also explored external authorization and automatic role management, but Jacob kept those separate and deferred statistics access while pursuing a common identity field and connection logging #22 #25 #28.
- Jacob's first complete series covered all authentication methods, logged the identity together with the authentication method and pg_hba.conf location, and included Kerberos-test and certificate-DN prerequisites, the latter extracted from Andrew Dunstan's work #29.
- Windows and BSD testing supported retaining compatibility-dependent SSPI identity formats, and the next revision removed the corresponding TODO without functional changes #30.
- Magnus's code review prompted removal of an unnecessary peer-user allocation and the log-escaping TODO, while Jacob explained that authentication must be logged even when authorization fails and that transport principals and mapped usernames cannot always replace the original identity #32 #33 #34 #35 #36.
- Michael Paquier proposed replacing position-tracking Kerberos log searches with restart-and-rotate isolation, then supplied an alternative using the logging collector; Jacob incorporated the restart approach into the complete series #39 #40 #41 #42.
- A simpler truncation-based Kerberos prerequisite landed, allowing Michael to remove that patch from the series; after briefly restoring ordered log matching, Jacob accepted independent matching within the isolated log fragment #43 #46 #47 #49.
- Michael's backend review led to explicit memory-lifetime documentation, a fatal error for invalid authentication-method values, and relocation of generic SCRAM password tests into the authentication suite #50 #51.
- Jacob defended recording identity before user-map authorization, and Michael accepted that distinction while requesting comments that would prevent future maintainers from moving the calls incorrectly #51 #52 #53.
- The certificate-DN prerequisite was absorbed by related SSL work, leaving one identity patch with explicit tests that full certificate DNs remain the authenticated identity under both CN and DN mapping #57.
- A prolonged test-design discussion weighed preserving diagnostic logs against restart overhead; Jacob introduced a LogCollector helper, but Michael replaced it with simpler checks and truncation because he considered the additional machinery unnecessary #54 #55 #58 #60.
- Jacob then identified false positives caused by matching earlier SSL connections, prompting agreement to centralize connection execution and log isolation in PostgresNode::connect_ok/fails #61.
- Jacob supplied two preparatory test migrations plus the feature patch, and Michael consolidated the prerequisites into a named-parameter interface for SQL and expected output rather than forwarding arbitrary psql options #62 #63.
- Jacob corrected lost Kerberos query checks and inactive SSL-test arguments, restored successful-connection stderr checks, and expanded LDAP identity coverage in a revised full series #64.
- Michael's subsequent rebase used his preceding preparatory patch, omitting Jacob's corrections, and landed as c50624cdd2 with light formatting and documentation edits; he then posted a remaining identity patch using an assertion for repeated authentication #65 #66.
- Jacob reported that the landing had disabled some Kerberos checks and that the remaining feature had lost his LDAP additions, so he rebased his earlier branch and separated the missing test corrections into a prerequisite #67.
- Michael confirmed that Jacob's messages had gone to spam and committed the Kerberos and SSL corrections as 5a71964a83, excluding successful-connection stderr checks to remain consistent with other suites #68.
- The final backend debate concerned repeated authentication: Jacob argued that a FATAL should immediately terminate the affected connection, and Michael accepted that reasoning instead of an assertion or propagated error status #67 #68 #69 #70.
- Jacob's final remaining patch landed as 9afffcb833 with light formatting and documentation edits, retaining centralized identity capture and log_connections output while leaving statistics exposure for future work #69 #70.
- Jacob subsequently reported that the buildfarm was green after test fixups, and Michael pointed to related follow-up work on other TAP-test paths #71 #72.
Sep 2021 · 3 entries
- Months later, Peter Eisentraut noticed that the test refactor had removed a Test::Builder::Level increment and proposed restoring it along with analogous missing increments so failures would identify the outer test call #73 #75.
- Jacob and Michael reviewed Peter's correction before it landed unchanged as 73aa5e0caf #76 #77 #78.
- The thread ended with the submitted feature and corrections landed; broader Test::Builder::Level coverage and documentation remained follow-up topics, which Andrew Dunstan raised and Michael proposed moving to a new thread #78 #79 #80.