Devel::MAT::UserGuide - a users' introduction to Devel::MAT
The Devel::MAT ecosystem allows developers of perl programs to inspect and analyse memory-related problems such as memory leaks, unexpected memory consumption, or odd state. This is an "offline" analysis system, in the sense that the analysis tools all run in a different process, possibly at a later time, than the perl process that is being analysed.
The basic workflow consists of two main stages: first a heap dump file is generated from the perl process being debugged, at or around the time that the problem becomes apparent, and secondly this file is loaded by an analysis tool in order to inspect the contents.
These two stages are described here separately. It is important to note that they don't have to be done at the same time, on the same perl process, or even on the same sort of machine. It is fully possible to capture the heap from a process on, say, a small server, then copy the file to a development workstation or laptop and analyse it there at a later time. It is for this reason that the heap-dumping part, Devel::MAT::Dumper, is now separated into its own CPAN distribution. This means it can be installed on its own, without all the extra dependencies the full set of analysis tools require.
To generate the heap dump file that captures the contents of the heap, the Devel::MAT::Dumper module is used. Ultimately the dump function within it needs to be called, but usually one of the module load options can be used on the perl commandline to achieve this without requiring the running code to be modified.
For example, the -dump_at_DIE option means that a heap dump will be written just before the process quits due to an uncaught exception:
$ perl -MDevel::MAT::Dumper=-dump_at_DIE program.pl
At this point, the program will start up and run normally, but if it is about to die, it will first write a .pmat file capturing the contents of the memory.
Dumping to program.pl.pmat because of DIE
Can't call method "method" on an undefined value at program.pl line 123.
There are a variety of other options for other situations, to suit other sorts of bugs and issues under investigation. For more options, see the documentation at "IMPORT OPTIONS" in Devel::MAT::Dumper.
Now that we have a .pmat file, we can load it and start to inspect the contents. A lot of the smaller, simpler tools are built as plugins for the main pmat command shell, so we can start by loading the heap file there.
$ pmat program.pl.pmat
Perl memory dumpfile from perl 5.24.1
Heap contains 15624 objects
In this shell a collection of commands is available to help analyse and inspect the contents of memory represented by this heap dump, which can be used in an interactive way, trying to narrow down to find the cause of the memory issue.
It is hard in general to describe exactly what sequence of analysis commands will be best to find the problem, as the specifics of each individual case will call for different kinds of analysis and require us to ask different questions of the toolset.
Ultimately there is quite a variety of possible underlying causes of memory growth in a Perl program; a few possible causes could be:
A single large SV such as a hash or array containing millions of items, or a single string possibly gigabytes in length.
A large number of SVs being created retained indefinitely, never being reclaimed.
A large number of temporary SVs being created, but due to internal reference cycles their memory is never reclaimed despite them now being unreachable.
This list of course is quite incomplete - there are as many different variations of erroenous memory usage as there are possible programs to write. Additionally, a lot of more interesting programs often suffer multiple overlapping issues at once. Nevertheless, this broad categorisation can help to describe some overall approaches to finding memory usage issues.
A good first step to take in the pmat shell to try to distinguish these cases is to use the largest command. This command requires no additional arguments, and by default will print (in size order), the five largest individual SVs in the entire heap.
For more information about the largest command, see also "largest" in Devel::MAT::Tool::Sizes.
Sometimes you'll find a single SV that far outweighs all the others; for example:
SCALAR(PV) at 0x6a47708: 1.6 GiB
SCALAR(PV) at 0x1a59488: 4.0 MiB
HASH(0) at 0xfb4770=strtab: 1.5 MiB
SCALAR(PV) at 0x71b6468: 707.3 KiB
SCALAR(PV) at 0x71be2f0: 609.6 KiB
others: 46.2 MiB
In this output, we see that the topmost SV reported, at address 0x6a47708 is much larger than everything else put together. In this case we have essentially already found the cause of the memory usage growth, and we can proceed by identifying what this particular SV actually is, by following the process in Devel::MAT::UserGuide::IdentifyingAnSV.
For a brief overview, we can just count the total number of objects of various kinds in the heap:
Kind Count (blessed) Bytes (blessed)
ARRAY 182 0 16.0 KiB
CODE 182 0 22.8 KiB
GLOB 325 0 48.2 KiB
We can inspect the callstack at the time the heap dump was made:
program.pl line 152: &main::func => void
$_: SCALAR(PV) at 0x55c2bdce2778 = "arguments"
$_: SCALAR(PV) at 0x55c2bdce2868 = "go"
$_: SCALAR(PV) at 0x55c2bdce26e8 = "here"
A list of the commands currently available in the shell can be found by the help command:
callstack - Display the call stack
count - Count the various kinds of SV
elems - List the elements of an ARRAY SV
For more information about a particular command, give its name as an argument to the help command:
pmat> help sizes
sizes - Summarize object and byte counts across different SV types
--owned sum SVs by owned size
--struct sum SVs by structural size
Also note that each command is implemented by a (correctly-cased) package under the Devel::MAT::Tool namespace. For example, the count tool is implemented by, and therefore more documentation can be found in, the Devel::MAT::Tool::Count package.
Many commands operate on a particular given SV. This can be specified in several ways:
A numerical address directly:
pmat> show 0x55a7e4e59f78
IO()=IO::File at 0x55a7e4e59f78 with refcount 1
A named root SV (see the roots command for a list of them all):
pmat> show defstash
STASH(61) at 0x55a7e4d69060=defstash with refcount 2
A named symbol from the symbol table. Note that subs require the & sigil:
pmat> show $warnings::VERSION
SCALAR(PV) at 0x55a7e4d96550 with refcount 1
pmat> show &warnings::import
CODE(PP) at 0x55a7e4dc3458 with refcount 1
Paul Evans <firstname.lastname@example.org>
To install Devel::MAT, copy and paste the appropriate command in to your terminal.
perl -MCPAN -e shell
For more information on module installation, please visit the detailed CPAN module installation guide.