Superior-Cache-ANalyzer 3.3.5

An analysis tool for Apache Traffic Server caches.

Superior Cache ANalyzer

F.A.Q

• Q: Scan says it can’t find my cache file, but I know it’s there. What do?

  A: Sometimes, ATS installs point to cache files that are installed relative to the ATS root directory. This is pretty common in test setups right after a basic install. There’s not really any way for SCAN to ‘detect’ when this is happening (though it will try), so the best solution is often just to try running scan from that ATS root directory. For example, if you have a directory /opt/trafficserver that holds all of the trafficserver files, try going to that directory before running scan.

• Q: Scan is giving me an error, and I don’t know what it means/how to fix it. How fix?

  A: Congratulations, you’ve just been drafted! Try running scan like this: scan --debug <other options that you specified last time> 2>scan.err (you may not see the error message this time) and then create a github issue and upload/pastebin/^C^V the scan.err file that should’ve been created and link/paste it into the box, along with a description of what you were trying to do and what went wrong. I’ll fix it as soon as I can.

• Q: Why can’t Scan see the thing that I KNOW is in the cache?

  A: It’s possible that you scaned for ‘thing’ before it was written. The Apache Traffic ServerTM will only sync directories every 60 seconds by default (effectively, this means scan can only see cache changes at that frequency). You could either wait a bit, or set the ATS configuration parameter proxy.config.cache.dir.sync_frequency to a lower value (in seconds). If that doesn’t work, check out the above question.

User Guide

SCAN’s primary use is as a library for inspecting Apache Traffic ServerTM (ATS) caches. SCAN also provides a command-line utility (scan), which is described here.

Installation

Prerequisites

scan requires the following dependencies:

• numpy - A highly-performant library for working with vectorized functions on huge data structures (used for reading/manipulating cache directories) link.

• psutil - A cross-platform process and system interface library (used for ionice setting) link.

• setuptools - “Easily download, build, install, upgrade, and uninstall Python packages” link.

• typing - Provides a backport of type-hinting for old versions of Python (< v3.5) link.

If you have Python version 3.5 or greater, you already have typing. If you have pip3 (for any Python version > 3.4.0), you likely already have setuptools. To install a dependency “DEP” on Python versions < 3.5, simply run sudo -H pip3 install DEP. If you don’t have pip3, then to install either dependency on CentOS/Fedora/RHEL distros, do sudo yum install -y python34-DEP, on Ubuntu/Mint/Debian distros do sudo apt-get install python3-DEP, and on Arch/Manjaro/Gentoo(?) do sudo pacman -S python3-DEP. If you need the dependencies and you’re on MacOS/BSD/Windows, then gods help you - I can’t.

Installing ‘scan’

Via pip

By far the easiest way to install SCAN is to simply use pip like so:

pip install Superior-Cache-ANalyzer

Note that you'll probably need to run that command as an
administrator (Windows), with ``sudo`` (Everything Else), or with
the ``--user`` option (Everything Including Windows)

From a Release

On the Releases page you can download the wheel (the .whl file) and install that manually with

sudo -H python3 -m pip install -y /path/to/Superior-Cache-ANalyzer.<version stuff>.whl

Note that this may require you to upgrade/install the ``pip``
module, so if you get an error like ``No module named 'pip'`` try
installing the ``python3-pip`` package (``python34-pip`` on
RedHat/CentOS/Fedora) or running ``sudo -H python3 -m ensurepip``.
Other errors could possibly be fixed by running
``sudo -H python3 -m pip install -yU pip`` and then trying the
install again. If all else fails, then you can probably install from
source.

From Source

To install from source, you’ll first want to download the source from the Comcast Github. Once you’ve done that, go to the downloaded folder and run

sudo -H pip3 install .

... or, if you don't have ``pip3``:

sudo python3 setup.py install

| Note that SCAN is only guaranteed to work for Python versions
  3.4.1 and greater.
| If you want to run the tests see 'Tests' below.

Usage

The basic usage of scan is pretty simple at the moment; to start the utility simply run:

scan [ --debug ] [ -f --fips ] [ -d --dump [ SPAN ] ] [ -c --config-dir DIR ]
scan [ --debug ] [ -f --fips ] [ -D --dump-breakdown [ SPAN ] ] [ -c --config-dir DIR ]

where the options have the following meanings:

• -c or --config-dir DIR

  This option allows you to directly specify the config dir of your ATS install. This allows you to skip the prompt when scan first starts where you must input your configuration directory. In non-interactive mode (-d/--dump given), this option must be used if ATS is not installed under /opt/trafficserver.

• --debug

  When provided, this flag causes scan to output some verbose debugging information and exception stack traces. It also causes it to be run without optimization, which - depending on your Python interpreter - can have a serious impact on performance.

• -d or --dump [SPAN]

  Dumps the contents of the cache in Tabular YAML format to stdout, then exits. This will cause any -D/--dupm-breakdown flags given to be ignored. If specified, SPAN should be the path to a cache span to dump as specified in storage.config e.g. /dev/sdk. WARNING: As of the time of this writing, scan’s “ionice” value is being set to the lowest possible value on startup, which means that this operation could take several hours to complete if you do not specify a single span. Currently, if you do not use the -l or --loadavg option, it takes about 400-500 seconds to dump a 1TB hard disk cache and about 3-7 seconds to dump an 8GB RAM cache. Use of this option with -l or --loadavg is not recommended at this time, as it will radically increase the time it takes to complete.

• -D or --dump-breakdown [SPAN]

  Dumps the usage of the cache to stdout in Tabular YAML format, broken down by host, then exits. If -d/--dump was given on the command line, this flag will be ignored if present. If specified, SPAN should be the path to a cache span to dump as specified in storage.config e.g. /dev/sdk. WARNING: As of the time of this writing, scan’s “ionice” value is being set to the lowest possible value on startup, which means that this operation could take several hours to complete if you do not specify a single span. Currently, if you do not use the -l or --loadavg option, it takes about 400-500 seconds to dump a 1TB hard disk cache and about 3-7 seconds to dump an 8GB RAM cache. Use of this option with -l or --loadavg is not recommended at this time, as it will radically increase the time it takes to complete.

• -f or --fips

  You must use this option if the ATS running on your system was compiled with ENABLE_FIPS enabled. If you don’t, everything will be messed up. Actually, some things will still be messed up even if you do.

• -l or --loadavg LOADAVG

  This flag allows the specification of a maximum system load average to be respected by the program. This is expected to be a comma-separated list of floating-point numbers (see `man uptime <https://linux.die.net/man/1/uptime>`__). For example: scan -l "25.0, 25.0, 25.0" ensures that no more than 25 processes will be waiting for CPU time or disk I/O on average ever 1, 5 or 15 minutes. Note that this option assumes that the system’s loadavg at the time scan starts is representative of the system’s loadavg for the entirety of its execution; if you start a very long scan job on e.g. a 1TB span, and then decide to play Crisis 1 on Medium settings using integrated graphics, your system may very well exceed a specified maximum loadavg, through no fault of scan itself. Note that if your system is already at or above the LOADAVG specified, scan will immediately exit as it cannot possibly run. (Implementation note: effectively this controls the number of sub-processes that can be used to scan a stripe at once, since each sub-process is potentially another process that will wait for CPU time or Disk I/O.) Note that this is only available on POSIX-compliant systems. Usage of this flag alongside -d or --dump is discouraged.

• -V or --version

  Prints the version information and exits. This will print both scan’s version and then on the next line the version and implementation of the Python interpreter used to run it. This second line would - for example - usually look like the follow on CentOS7.x systems: Running on CPython v3.4.5.

Once the utility is started (provided the -d/--dump or -c/--config flags are not given) you’ll be faced with a pretty basic prompt. At first, your only option will be [1] Read Storage Config. After you select this option, you’ll be prompted to enter the location of your ATS configuration files. “Tab-completion” is supported for most interactive prompts, including the ATS configuration file prompt. SCAN will expect all of them to be in the same directory, and will guess that they are in /opt/trafficserver/etc/trafficserver/ by default. Note that the use of FIPS at compilation time cannot be determined from the config files, and MUST be given on the command line. Once the configuration has been read, all menu options will be unlocked. They are as follows:

[1] Show Cache Setup

This option will print out the spans and volumes declared in the configuration. Output will look like:

Cache files:
/path/to/a/span Span of <n> stripes XXX.XB

Volumes:
#1  <type>  XXX.XB

where <n> is the number of stripes in the span on that line, and XXX.XB is the size of a span/volume (but it will be displayed in human-readable approximations in units of B, kB, MB, or GB as appropriate). Volumes defined as a percent of total storage will have their size calculated at runtime, and displayed in absolute terms. <type> will be the type of volume declared. In nearly all cases, this will be http, but certain plugins could define other volume types. Finally, it should be noted that while this example shows one volume on one span, this menu option will display all volumes and all spans, in no particular order and with no distinction between cache spans on files, block devices, or ram devices.

[2] List Settings

This option will list the settings declared in records.config, in proper ATS syntax. An example:

proxy.config.log.collation_host STRING NULL
proxy.config.ssl.compression INT 1

Only one or two of these settings actually has any impact on the function of scan, but all values are read in to facilitate future extension.

[3] Search for Setting

This option will bring up a prompt to type a search string for a specific setting from records.config. Python-syntax regex is supported and enabled by default (meaning searching for ‘proxy.config’ will match ‘proxyZconfig’ as well as the exact string typed).

[4] List Stripes in a Span

This option will prompt you to enter a span (which is the full path to the span file) and then list all stripes within it. The output is in the format:

XXX.XB stripe, created Www Mmm D hh:mm:ss (version XX.X)

where XXX.XB is the size of a stripe (but it will be displayed in human-readable approximations in units of B, kB, MB, or GB as appropriate), Www Mmm D hh:mm:ss is the date of the stripe’s creation (in the system’s ctime(3) format) and XX.X is the decimal-separated major and minor version numbers of the cache system that created it. Note that this version is not the same as the version of ATS using the cache. Also note that as of this time only version 24.0+ is supported by scan, and using lower versions with scan will cause to crash and/or give incorrect output.

[5] View URLs of objects in a Span

When selected, this option will first prompt you for a span. It will then search all of the stripes on that span for stored objects, and catalog their URLs, printing them to the screen as they are found. Each URL is printed in the format:

protocol://[[user]:password@]host/path/to/content    - XXX.XB -     x<Y>

where protocol is the protocol used to retrieve the content (nearly always http or https), [[user]:password@] is the username (if used, usually not) ‘colon’ password (if used, usually not) used to access the content ‘at’ the host - which is the fully-qualified domain name of the content host, and path/to/content is the location on that host of the content stored in the cache. A typical example of a path is images/test/testquest.png. XXX.XB is the size of this content (but it will be displayed in human-readable approximations in units of B, kB, MB, or GB as appropriate). Finally, <Y> will be the number of times this same URL is stored in the cache (typically in ‘alternate’ forms). For example, if a given item is stored only once in the cache span, its line will end in x1, and if it is encountered 42 times, then it will end in x42. Note that the size of a given object is reported as the size of one instance of this item, regardless of the number actually stored.

Warning: When tested on a span of a single, roughly 830GB stripe, this operation took between 39 and 44 seconds to complete. Be aware that the time this takes is directly proportional to the size of the spans, and the number of spans that it is searching. However, results are cached so that subsequent searches (or uses of menu option 6) on the same span should be significantly quicker. To help recognize that the program has not frozen, findings are printed to the screen as they are found, and the main menu will display upon completion.

[6] View Usage of a Span broken down by host

This option will first prompt for a span, then it will list the hosts that have content stored in that span, as well as the total storage size used, the storage size as a percent of the total available storage, and the storage size as a percent of the storage currently in use. The output format for each host is as follows:

<host>   - XXX.XB -     YY.YY% of available space -     ZZ.ZZ% of used space

where <host> is the fully-qualified domain name of the host, XXX.XB is the total size of that host’s content on disk (but it will be displayed in human-readable approximations in units of B, kB, MB, or GB as appropriate), YY.YY is the percent of available space taken up by this host’s content, and ZZ.ZZ is the percent of space currently being used to store objects that is taken up by this host’s content.

Warning: When tested on a span of a single, roughly 830GB stripe, this operation took between 39 and 44 seconds to complete. Be aware that the time this takes is directly proportional to the size of the spans, and the number of spans that it is searching. However, results are cached so that subsequent searches (or uses of menu option 5) on the same span should be significantly quicker. To help recognize that the program has not frozen, findings are printed to the screen as they are found, and the main menu will display upon completion.

[7] Dump cache usage stats to file (Tabular YAML format)

This option will ask you to first name a file for output (relative or absolute paths - doesn’t matter which), then it will dump the output of a call to the ‘View URLs of objects in a Span’ for ALL spans in the cache system to the named file in Tabular YAML (TYAML) format (which is just YAML but indented with tabs instead of spaces and accepts None as a null value.)

Tests

If you want to run the tests, be sure you’re in the project’s root directory and run the test.sh script. Note that the unit tests will download and attempt to build Apache Traffic Server from source and as such will also require all of the dependencies of Apache Traffic Server. A minimal linting test (Good for auditing your contribution at a glance) can be run with pylint by just running pylint --rcfile=./.pylintrc scan/ from the project’s root directory. A pylint score above 9.5 and with no erros (e.g. E001: SyntaxError) is considered “passing”.

Tabular YAML Format

The output of the interactive mode’s 7th option and the -d or --dump option are given in what’s been referred to as “Tabular YAML Format”. As the name implies, this is similar to YAML. In fact, it should be considered syntactically identical to YAML but for one exception: indentation is always done via the tab character, **never with spaces**. This was done because without harming its human readability, it allows for much easier pipelining of output e.g. via cut.