Skip to main content

Audit 1.12.0

Tera Insights Team
Tera Insights Team
tiCrypt Team

What is in this release​


Changes​

  • A log line that log-uploader or the log streamer cannot load is reported once, with its error, the record it was read as and the text of the line, quoted and cut to 600 characters. Before, the message named neither the record nor the line, and a file that was not a log was printed as it was, which could garble the terminal. The verbose mode of log-uploader no longer lists the failed lines again at the end
  • A query that the audit database ends part-way, because it ran past the time limit or needed too much memory, fails with that reason: choose a shorter date range and run it again. Before, it could show the rows found so far as if they were the whole result
  • The SQL workbench is read-only: a statement that writes to the audit database or changes one of its tables is refused. Before, such a statement was refused on only one of the ten database connections tiAudit uses
  • When a query result is too large for the web view, a notice above the table says how many rows are shown and how many there are, for example "Showing the first 10,000 of 2,630,281 rows". Before, the result was cut without notice, and the viewer always showed "Query results in browser may be truncated"
  • The values given for a query's parameters can no longer alter the query: text is taken as it is, quotes included, and a result size or other number must be a whole number. Before, a logged-in user could add conditions to a built-in query, or read other tables through it, by putting SQL in a parameter or in the end date
  • log-uploader does not load log lines the audit database already has: lines it loaded before, and lines the log streamer stored. Before, loading a log file a second time stored every row twice. Use --load-again to load them anyway
  • When a query result is too large for the web view, the export menu reads "Export the 10,000 rows shown", and the exported file says on its Summary sheet how many rows it holds and how many the query returned. The sheet with the rows is named "Results (first 10,000 rows)". Before, the entry read "Export all data" and nothing in the file showed that rows were missing
  • "Projects Members", "Project Members", "Project Statistics", the "#Project Members" column of "Projects Listed", "List user projects" and the project count of "Users Listed" leave out project memberships that have expired, and follow later changes to a membership's expiration. Before, a user stayed listed as a member after the membership expired, because tiCrypt writes no log record when one expires

Fixes​

  • The log streamer waits for ClickHouse and loads the lines once it is back. Before, lines that arrived while ClickHouse was restarting, out of disk space or refusing inserts were dropped. log-uploader tries a failed batch again for up to two minutes
  • The log streamer no longer stops loading after ten inserts fail, for example into a table that is missing. Before, each failed insert kept a database connection, and the streamer loaded nothing more until it was restarted
  • When the connection from tiCrypt breaks and tiCrypt sends its last block of lines again, the log streamer loads each line once. Before, the lines were loaded twice and a hash chain alert was raised over an intact log
  • A row that ClickHouse refuses is left out and reported, and the rest of its batch is loaded. Before, log-uploader lost the whole batch, up to 1000 lines
  • A log line whose timestamp is not a valid time is reported as a failed line. Before, it stopped log-uploader and made the log streamer restart
  • VM proxy connections and libvirt hosts with a port above 32767 are loaded. Before, the line failed with "value out of range". The DestPort and NatPort columns of existing tables are converted the first time the new version starts
  • An alert is sent to every configured alert driver when one of them fails, and a Splunk server that does not answer no longer holds up the log streamer
  • log-uploader reads the debugQueries setting from its configuration file
  • The log streamer and log-uploader start on an audit database that was created before version 1.9.0, and load into its tables as they are. Before, they stopped at startup with "the type of column ... does not match database", and the only way on was to delete the database and load every log file again
  • An escrow recovery key record with no recovery key ID is loaded. Before, it was refused, and the records loaded in the same batch were lost with it
  • tiAudit no longer stops answering when four or more lists of projects are loaded at the same time. Before, every query, report and page waited until the tiAudit service was restarted
  • The "#Project Members" column of "Projects Listed" counts the users who are in the project now. Before, a user who had been removed from any project was left out of every project, so the counts were too low
  • "List user groups", "List user permissions", "List user projects" and "List user teams" say that the user is not in the audit log as of the end date. Before, they failed with "Could not extract Groups from <nil>"
  • An upload started from the upload page no longer stops for good when log-uploader prints a line longer than 64 KB. Before, the upload stayed "Running" until tiAudit was restarted, with the audit database only partly loaded. A line of output over 16 KB is shown cut
  • Records with a number above 32767 in a team permission mask, a set of key flags, a key index, a protocol version, or a count of attachments, hosts, virtual cores or cleared drives are loaded. Before, the line failed with "value out of range". These columns of existing tables are converted when the log streamer or log-uploader starts
  • When an upgrade adds two neighbouring columns to a log table, both are added at the first start of the log streamer or log-uploader. Before, one of them could be left out, and the records of that table were refused with "No such column" until the next start
  • "Libvirt Host Count" and the "VM host added" rows of "Privileged Actions Audit" count only records that describe a host. Before, logs from 2019 and 2020 added thousands of host additions that never happened
  • log-uploader reads the files of a log directory in the order tiCrypt wrote them, the current file last. Before, it read the current file first, and every load raised a hash chain alert over an intact log unless the directory still held the first log file
  • Two runs of the same report at the same time no longer share their dates and their user, project or team. Before, one report could come out with the parameters of the other, tiAudit could stop, and the CSV export of a custom report whose query has no parameters failed with "assignment to entry in nil map"
  • Times in exported reports are written year-month-day. Before, the XLSX and CSV exports and the automatic reports wrote year-day-month, so 3 April 2026 read as 2026-03-04
  • A query with a very large result no longer makes tiAudit hold the whole result in memory for the web view: the rows after the 10,000 shown are counted and not kept. Before, a result of 2.6 million rows took about 600 MB until it was answered
  • "List user projects" and the project count of "Users Listed" include a project the user was removed from and added to again on the same day. Before, such a project was left out