Make a dump of the V8 heap for later inspection.
npm install heapdump
Or, if you are running node.js v0.6 or v0.8:
npm install [email protected]
node-gyp configure build
Load the add-on in your application:
var heapdump = require('heapdump');
The module exports a single writeSnapshot([filename], [callback])
function
that writes out a snapshot. filename
defaults to heapdump-<sec>.<usec>.heapsnapshot
when omitted.
You can specify NODE_HEAPDUMP_FILENAME
env variables, which will be used as template for
filename (include folder) - NODE_HEAPDUMP_FILENAME="/var/heapdumps/heapdump-{sec}.{usec}.snapshot"
,
in case if you want to save snapshots in different folder than application's working directory.
heapdump.writeSnapshot('/var/local/' + Date.now() + '.heapsnapshot');
The function also takes an optional callback function which is called upon completion of the heap dump.
heapdump.writeSnapshot(function(err, filename) {
console.log('dump written to', filename);
});
The snapshot is written synchronously to disk. When the JS heap is large, it may introduce a noticeable "hitch".
On UNIX platforms, you can force a snapshot by sending the node.js process a SIGUSR2 signal:
$ kill -USR2 <pid>
The SIGUSR2 signal handler is enabled by default but you can disable it
by setting NODE_HEAPDUMP_OPTIONS=nosignal
in the environment:
$ env NODE_HEAPDUMP_OPTIONS=nosignal node script.js
Open Google Chrome and press F12 to open the developer toolbar.
Go to the Memory
tab, right-click in the tab pane and select
Load profile...
.
Select the dump file and click Open
. You can now inspect the heap snapshot
at your leisure. Some snapshots may take a long time to load, on the order of
minutes or even hours.
Note that Chrome will refuse to load the file unless it has the .heapsnapshot
extension.
On UNIX systems, the rule of thumb for creating a heap snapshot is that it
requires memory twice the size of the heap at the time of the snapshot.
If you end up with empty or truncated snapshot files, check the output of
dmesg
; you may have had a run-in with the system's OOM killer or a resource
limit enforcing policy, like ulimit -u
(max user processes) or ulimit -v
(max virtual memory size).