View counter 6.0.15: multi-site logs, resume state and what counts as a view #

Read this page if you run the cronjob updateviewcount.php, or if templates or extensions show view_count or use the fetch content/view_top_list. The cronjob was rebuilt for 6.0.15 so that installations with several sites can count their views correctly; the guide View count statistics explains why and how to set it up.

In short #

Behavior that changed #

• The resume state is no longer var/<site>/log/updateview.log with the last line read. It is var/<site>/viewcount/<log>-<hash>.json ([ViewCountSettings] StateDir), one file per log, with each siteaccess's inode and byte offset in it, and a .lock file next to it. The first run on a log without state takes over the old file (updateview-<log name>.log, else updateview.log): it starts after the line recorded there, or, when that line is gone, counts only newer lines. The old file is left in place and no longer written; it can be removed after the first run.
• "/" counts for the node of the siteaccess's site.ini [SiteSettings] IndexPage, no longer for the tree root (or a PathPrefix's top node). An IndexPage that is no node (user/login) means "/" is not counted.
• A path with a PathPrefix is looked up with each prefix first, then as it is; PathPrefixExclude[] is respected. With logfile.ini PathPrefix[] empty, the siteaccess's own PathPrefix is used.
• SitePrefix[] removes a whole first path element only (mysite no longer applies to /mysite2/...).
• Only page views count: GET with status 200 or 304, no robots (BotUserAgents[], an empty user agent), no files (SkipExtensions[]), no module and service paths (SkipPathPrefixes[]). The earlier cronjob counted every line whose path was content, whatever its method, status or user agent, so counts are lower for the same traffic. Empty lists restore the earlier behavior for that check.
• /content/view/<view mode>/<node> counts for every view mode, not only full (ViewModes[] limits it), and view parameters (/(offset)/10) and URL encoding (%C3%BC) no longer make a path miss its node.
• Several counted siteaccesses reading one log get each line once, by the kernel's siteaccess matching; the earlier cronjob counted every line for every siteaccess run.
• Rotation: the rest of the rotated copy is read before the new log (RotatedSuffixes[]); a truncated log is read from its start with a warning.
• One transaction per run: all views of a run are written together, and the state only after the commit. Nodes that do not exist are left out with one query per 500 nodes instead of one per node.
• Lines longer than 1023 bytes are read whole (up to MaxLineLength, default 64 KiB).
• Output: the cronjob prints a report per log (lines read, views, reasons lines were not counted, the state) instead of one line per node.
• The cronjob's description no longer speaks of an "in-memory view count buffer"; none exists. Setup > Cronjobs shows the new text.

New API #

• eZViewCounter::addCounts( array $counts ): node id => views, added in one transaction (count = count + n for rows that exist, new rows for the rest); returns false when the database refused the transaction.
• eZSiteAccess::matchAddress( $uriString, $host, $port = 80 ): the siteaccess that served an address, as match() finds it, without the redirect of RedirectOnNormalize and without the browser's language; unmatched is set where only DefaultAccess applied.
• expViewCountUpdater, expViewCountConfiguration, expViewCountLogReader, expViewCountLogParser, expViewCountRequestFilter, expViewCountPathResolver, expViewCountState, expViewCountSite: the counter's parts, each usable on its own (an extension can count from another source with its own expViewCountSite list).

Credits #

The per-log resume file, the front page counted for the siteaccess's IndexPage and the fallback to the path without PathPrefix come from a prototype a site operator wrote for a multi-site installation.