impstats: Generate Periodic Statistics of Internal Counters¶
|Author:||Rainer Gerhards <firstname.lastname@example.org>|
This module provides periodic output of rsyslog internal counters.
The set of available counters will be output as a set of syslog messages. This output is periodic, with the interval being configurable (default is 5 minutes). Be sure that your configuration records the counter messages (default is syslog.=info). Besides logging to the regular syslog stream, the module can also be configured to write statistics data into a (local) file.
When logging to the regular syslog stream, impstats records are emitted just like regular log messages. As such, counters increase when processing these messages. This must be taken into consideration when testing and troubleshooting.
Note that loading this module has some impact on rsyslog performance. Depending on settings, this impact may be noticeable for high-load environments, but in general the overhead is pretty light.
Note that there is a rsyslog statistics online analyzer available. It can be given a impstats-generated file and will return problems it detects. Note that the analyzer cannot replace a human in getting things right, but it is expected to be a good aid in starting to understand and gain information from the pstats logs.
The rsyslog website has an overview of available rsyslog statistic counters. When browsing this page, please be sure to take note of which rsyslog version is required to provide a specific counter. Counters are continuously being added, and older versions do not support everything.
The configuration parameters for this module are designed for tailoring the method and process for outputting the rsyslog statistics to file.
Parameter names are case-insensitive.
This module supports module parameters, only.
Sets the interval, in seconds at which messages are generated. Please note that the actual interval may be a bit longer. We do not try to be precise and so the interval is actually a sleep period which is entered after generating all messages. So the actual interval is what is configured here plus the actual time required to generate messages. In general, the difference should not really matter.
The numerical syslog facility code to be used for generated messages. Default is 5 (syslog). This is useful for filtering messages.
The numerical syslog severity code to be used for generated messages. Default is 6 (info).This is useful for filtering messages.
When set to “on”, counters are automatically reset after they are emitted. In that case, the contain only deltas to the last value emitted. When set to “off”, counters always accumulate their values. Note that in auto-reset mode not all counters can be reset. Some counters (like queue size) are directly obtained from internal object and cannot be modified. Also, auto-resetting introduces some additional slight inaccuracies due to the multi-threaded nature of rsyslog and the fact that for performance reasons it cannot serialize access to counter variables. As an alternative to auto-reset mode, you can use rsyslog’s statistics manipulation scripts to create delta values from the regular statistic logs. This is the suggested method if deltas are not necessarily needed in real-time.
New in version 8.16.0.
Specifies the format of emitted stats messages. The default of “legacy” is compatible with pre v6-rsyslog. The other options provide support for structured formats (note the “cee” is actually “project lumberjack” logging).
The json-elasticsearch format supports the broken ElasticSearch JSON implementation. ES 2.0 no longer supports valid JSON and disallows dots inside names. The “json-elasticsearch” format option replaces those dots by the bang (“!”) character. So “discarded.full” becomes “discarded!full”. Options: json/json-elasticsearch/cee/legacy
This is a boolean setting specifying if data should be sent to the usual syslog stream. This is useful if custom formatting or more elaborate processing is desired. However, output is placed under the same restrictions as regular syslog data, especially in regard to the queue position (stats data may sit for an extended period of time in queues if they are full).
If specified, statistics data is written to the specified file. For robustness, this should be a local file. The file format cannot be customized, it consists of a date header, followed by a colon, followed by the actual statistics record, all on one line. Only very limited error handling is done, so if things go wrong stats records will probably be lost. Logging to file an be a useful alternative if for some reasons (e.g. full queues) the regular syslog stream method shall not be used solely. Note that turning on file logging does NOT turn off syslog logging. If that is desired log.syslog=”off” must be explicitly set.
Binds the listener to a specific ruleset.
New in version 8.4.1.
This is a utility setting for folks who post-process impstats logs and would like to know the begin and end of a block of statistics. When “bracketing” is set to “on”, impstats issues a “BEGIN” message before the first counter is issued, then all counter values are issued, and then an “END” message follows. As such, if and only if messages are kept in sequence, a block of stats counts can easily be identified by those BEGIN and END messages.
Note well: in general, sequence of syslog messages is not strict and is not ordered in sequence of message generation. There are various occasion that can cause message reordering, some examples are:
- using multiple threads
- using UDP forwarding
- using relay systems, especially with buffering enabled
- using disk-assisted queues
This is not a problem with rsyslog, but rather the way a concurrent world works. For strict order, a specific order predicate (e.g. a sufficiently fine-grained timestamp) must be used.
As such, BEGIN and END records may actually indicate the begin and end of a block of statistics - or they may not. Any order is possible in theory. So the bracketing option does not in all cases work as expected. This is the reason why it is turned off by default.
However, bracketing may still be useful for many use cases. First and foremost, while there are many scenarios in which messages become reordered, in practice it happens relatively seldom. So most of the time the statistics records will come in as expected and actually will be bracketed by the BEGIN and END messages. Consequently, if an application can handle occasional out-of-order delivery (e.g. by graceful degradation), bracketing may actually be a great solution. It is, however, very important to know and handle out of order delivery. For most real-world deployments, a good way to handle it is to ignore unexpected records and use the previous values for ones missing in the current block. To guard against two or more blocks being mixed, it may also be a good idea to never reset a value to a lower bound, except when that lower bound is seen consistently (which happens due to a restart). Note that such lower bound logic requires resetCounters to be set to off.
The impstats plugin gathers some internal statistics. They have different names depending on the actual statistics. Obviously, they do not relate to the plugin itself but rather to a broader object – most notably the rsyslog process itself. The “resource-usage” counter maintains process statistics. They base on the getrusage() system call. The counters are named like getrusage returned data members. So for details, looking them up in “man getrusage” is highly recommended, especially as value may be different depending on the platform. A getrusage() call is done immediately before the counter is emitted. The following individual counters are maintained:
utime- this is the user time in microseconds (thus the timeval structure combined)
stime- again, time given in microseconds
openfiles- number of file handles used by rsyslog; includes actual files, sockets and others
- This module MUST be loaded right at the top of rsyslog.conf, otherwise stats may not get turned on in all places.
Load module, send stats data to syslog stream¶
This activates the module and records messages to /var/log/rsyslog-stats in 10 minute intervals:
module(load="impstats" interval="600" severity="7") # to actually gather the data: syslog.=debug /var/log/rsyslog-stats
Load module, send stats data to local file¶
Here, the default interval of 5 minutes is used. However, this time, stats data is NOT emitted to the syslog stream but to a local file instead.
module(load="impstats" interval="600" severity="7" log.syslog="off" # need to turn log stream logging off! log.file="/path/to/local/stats.log")
Load module, send stats data to local file and syslog stream¶
Here we log to both the regular syslog log stream as well as a file. Within the log stream, we forward the data records to another server:
module(load="impstats" interval="600" severity="7" log.file="/path/to/local/stats.log") syslog.=debug @central.example.net
Explanation of output¶
Example output for illustration:
Sep 17 11:43:49 localhost rsyslogd-pstats: imuxsock: submitted=16 Sep 17 11:43:49 localhost rsyslogd-pstats: main Q: size=1 enqueued=2403 full=0 maxqsize=2
All objects are shown in the results with a separate counter, one object per line.
Line 1: shows details for
imuxsock, an object
submitted=16, a counter showing that 16 messages were received by the imuxsock object.
Line 2: shows details for the main queue:
main Q, an object
size, messages in the queue
enqueued, all received messages thus far
full, how often was the queue was full
maxqsize, the maximum amount of messages that have passed through the queue since rsyslog was started
Help with configuring/using