v0.11.2, OSX 10.8, CodeRange::GetNextAllocationBlock Allocation failed

[medikoo]

I have a heavy configuration script, that generates a lot of files.
It works succesfully on v0.8.23 and v0.10.8 but on v0.11.2 (all 64-bit) at some point It throws:

FATAL ERROR: CodeRange::GetNextAllocationBlock Allocation failed - process out of memory

I've tried few times, and everytime I get this error.

When I look in Activity Monitor I see that Node process takes around 500-600 MB during that, and that's as far I know not close to process memory limit.

[bnoordhuis]

That's a V8 error, I'm not sure there's much we can do about it. Do you have a test case I can try?

[medikoo]

@bnoordhuis I prepared best test case I could: http://medyk.org/node-5590-test-case.zip

Unpack and run npm run setup in main folder.
It's both memory and CPU intensive script.

On v0.11.2 I always get:

$ npm ru setup
npm WARN package.json [email protected] No readme data.

> [email protected] setup /Users/medikoo/Projects/eregistrations/node-5590-test-case
> node node_modules/mano/bin/setup

Load schemas... loaded in 8.84s
Load tags... loaded in 62.92s
Generate client-side models...FATAL ERROR: CodeRange::GetNextAllocationBlock Allocation failed - process out of memory
npm ERR! weird error 5
npm ERR! not ok code 0

and on v0.10.8 it's fine:

$ npm ru setup
npm WARN package.json [email protected] No readme data.

> [email protected] setup /Users/medikoo/Projects/eregistrations/lomas-tmp
> node node_modules/mano/bin/setup

Load schemas... loaded in 4.72s
Load tags... loaded in 32.31s
Generate client-side models... generated in 227.12s
All cool

While I was observing memory consumption, it looks that on v0.10.8 script oscillates in 500-650 MB while in v0.11.2 it reaches at some points 950MB and usually when it gets to 950MB it crashes (shouldn't it allow to go up to 1.4 GB ?)

I think it might be related to how garbage collector works and it looks that V8 that comes with v0.10.8 does better job at clearing memory on the way.

So questions that comes to me, are:

1. Is it ok that script crashes at around 1 GB memory usage (not 1.4 GB) ?
2. Shouldn't garbage collector prevent such crashes if there's memory that can be freed ?

[medikoo]

It's cleary a V8 thing, so I'm closing that.
I found that default limit is 1GB on 64bit (and that can be increased by --max-old-space-size option), so now it's clear why it crashes.

The only question is whether it's ok that it crashes if memory can be freed, but that's question for V8 not node.

[medikoo]

I've moved discussion to V8 -> https://code.google.com/p/v8/issues/detail?id=2703&thanks=2703&ts=1369905364