Haproxy log analyzer that tries to gives an insight of what's going on
HAProxy log analyzer
This Python package is a HAProxy log parser that allows you to analyze your HAProxy log files in multiple ways (see commands section below).
Currently only the HTTP log format is supported.
The current --help looks like this:
usage: haproxy_log_analysis [-h] [-l LOG] [-s START] [-d DELTA] [-c COMMAND] [-f FILTER] [-n] [--list-commands] [--list-filters] Analyze HAProxy log files and outputs statistics about it optional arguments: -h, --help show this help message and exit -l LOG, --log LOG HAProxy log file to analyze -s START, --start START Process log entries starting at this time, in HAProxy date format (e.g. 11/Dec/2013 or 11/Dec/2013:19:31:41). At least provide the day/month/year. Values not specified will use their base value (e.g. 00 for hour). Use in conjunction with -d to limit the number of entries to process. -d DELTA, --delta DELTA Limit the number of entries to process. Express the time delta as a number and a time unit, e.g.: 1s, 10m, 3h or 4d (for 1 second, 10 minutes, 3 hours or 4 days). Use in conjunction with -s to only analyze certain time delta. If no start time is given, the time on the first line will be used instead. -c COMMAND, --command COMMAND List of commands, comma separated, to run on the log file. See -l to get a full list of them. -f FILTER, --filter FILTER List of filters to apply on the log file. Passed as comma separated and parameters within square brackets, e.g ip[192.168.1.1],ssl,path[/some/path]. See --list- filters to get a full list of them. -n, --negate-filter Make filters passed with -f work the other way around, i.e. ifthe ``ssl`` filter is passed instead of showing only ssl requests it will show non-ssl traffic. If the ``ip`` filter isused, then all but that ip passed to the filter will be used. --list-commands Lists all commands available. --list-filters Lists all filters available.
Commands are small purpose specific programs in itself that report specific statistics about the log file being analyzed. See the --help (or the section above) to know how to run them.
- Reports how many log lines could be parsed.
- Reports how many log lines could not be parsed.
- Reports a breakdown of how many requests have been made per HTTP method (GET, POST…)
- Reports a breakdown of how many requests have been made per IP. Note that for this to work you need to configure HAProxy to capture the header that has the ip on it (usually the X-Forwarded-For header). Something like: capture request header X-Forwarded-For len 20
- Reports the 10 IPs with most requests (and the amount of requests).
- Reports a breakdown of how many requests per HTTP status code (404, 500, 200, 301..) are on the log file.
- Reports a breakdown of how many requests per path (/rss, /, /another/path).
- Reports the 10 paths with most requests.
- Reports a list of requests that downstream servers took more than 1 second to response.
- Reports the amount of requests that downstream servers took more than 1 second to response.
- Reports the average time (in milliseconds) servers spend to answer requests. .. note:: Aborted requests are not considered.
- Reports the average time (in milliseconds) requests spend waiting on the various HAProxy queues.
- Reports a breakdown of how many requests were processed by each downstream server. Note that currently it does not take into account the backend the server is configured on.
- Reports a list of queue peaks. A queue peak is defined by the biggest value on the backend queue on a series of log lines that are between log lines without being queued.
- Reports on how many requests were made on SSL and how many on plain HTTP. This command only works if the default port for SSL (443) appears on the path.
- Reports on how many requests were made per minute. It works best when used with -s and -d command line arguments, as the output can be huge.
- Prints the raw lines. This can be useful to trim down a file (with -s and -d for example) so that later runs are faster.
Filters, contrary to commands, are a way to reduce the amount of log lines to be processed.
The -n command line argument allows to reverse filters output.
This helps when looking for specific traces, like a certain IP, a path…
- Filters log lines by the given IP.
- Filters log lines by the given IP range (all IPs that begin with the same prefix).
- Filters log lines by the given string.
- Filters log lines that are from SSL connections. See :method::.HaproxyLogLine.is_https for its limitations.
- Filters log lines that take at least the given time to get answered (in milliseconds).
- This is an implicit filter that is used when --start, and optionally, --delta are used. Do not type this filter on the command line, use --start and --delta.
- Filters log lines that match the given HTTP status code (i.e. 404, 200…).
- Filters log lines that match the given HTTP status code family (i.e. 4 for all 4xx status codes, 5 for 5xx status codes…).
- Filters log lines by the HTTP method used (GET, POST…).
- Filters log lines by the HAProxy backend the connection was handled with.
- Filters log lines by the HAProxy frontend the connection arrived from.
- Filters log lines by the downstream server that handled the connection.
- Filters log lines by the response size (in bytes). Specially useful when looking for big file downloads.
After installation you will have a console script haproxy_log_analysis:
$ python setup.py install
- add more commands: (help appreciated)
- reports on servers connection time
- reports on termination state
- reports around connections (active, frontend, backend, server)
- your ideas here
- think of a way to show the commands output in a meaningful way
- be able to specify an output format. For any command that makes sense (slow requests for example) output the given fields for each log line (i.e. acceptance date, path, downstream server, load at that time…)
- your ideas
- Fix issue #9. log line on the syslog part was too strict, it was expecting the hostname to be a string and was failing if it was an IP. [GF]
- Finally really fixed issue #7. namespace_packages was not meant to be on setup.py at all. Silly copy&paste mistake. [GF]
- Fix release on PyPI. Solves GitHub issue #7. https://github.com/gforcada/haproxy_log_analysis/issues/7 [GF]
- Fix release on PyPI (again). [GF]
- Fix release on PyPI. [GF]
- Pickle :class::.HaproxyLogFile data for faster performance. [GF]
- Add a way to negate the filters, so that instead of being able to filter by IP, it can output all but that IP information. [GF]
- Add lots of filters: ip, path, ssl, backend, frontend, server, status_code and so on. See --list-filters for a complete list of them. [GF]
- Add :method::.HaproxyLogFile.parse_data method to get data from data stream. It allows you use it as a library. [bogdangi]
- Add --list-filters argument on the command line interface. [GF]
- Add --filter argument on the command line interface, inspired by Bogdan’s early design. [bogdangi] [GF]
- Create a new module :module::haproxy.filters that holds all available filters. [GF]
- Improve :method::.HaproxyLogFile.cmd_queue_peaks output to not only show peaks but also when requests started to queue and when they finsihed and the amount of requests that had been queued. [GF]
- Show help when no argument is given. [GF]
- Polish documentation and docstrings here and there. [GF]
- Add a --list-commands argument on the command line interface. [GF]
- Generate an API doc for HaproxyLogLine and HaproxyLogFile. [bogdangi]
- Create a console_script haproxy_log_analysis for ease of use. [bogdangi]
- Add Sphinx documentation system, still empty. [GF]
- Keep valid log lines sorted so that the exact order of connections is kept. [GF]
- Add quite a few commands, see README.rst for a complete list of them. [GF]
- Run commands passed as arguments (with -c flag). [GF]
- Add a requirements.txt file to keep track of dependencies and pin them. [GF]
- Add travis and coveralls support. See its badges on README.rst. [GF]
- Add argument parsing and custom validation logic for all arguments. [GF]
- Add regular expressions for haproxy log lines (HTTP format) and to parse HTTP requests path. Added tests to ensure they work as expected. [GF]
- Create distribution. [GF]