On Mon, May 18, 2009 at 12:06:50PM +0800, Jerry wrote:
> 
> I just used checking tool to scan ganglia code and found 2 defects.

which tool?, did you check against the sources of ganglia-3.1.1?

> Do you think this can explain the 3G virtual memory issue? Below is
> detail of these 2 defects.

unlikely, 3GB is a lot of memory and those examples refer to really
small structures while the code around it hasn't changed that much since 3.0

> gmetad/gmetad.c, line 149
> 
> 146      if (do_sum)
> 147         {
> 148  tt = in_type_list(type, strlen(type));
> 149            if (!tt) return 0;
> 
> If we returned on line 149, then storage pointed by "hash_datum" will leak.

this will only happen if gmetad is told to summarize a type that it can't
identify (hence very unlikely to happen unless you have types other than
"int8", "uint8", "string", "timestamp", "int16", "uint16", "int32", "uint32",
"float", "double" in your XML).

in any case, a fix for this Committed revision 1997 for trunk.

> apr-1.3.3/memory/unix/apr_pools.c, line 946
> 
> 941       if ((node = allocator_alloc(pool_allocator,
> 942MIN_ALLOC - APR_MEMNODE_T_SIZE)) == NULL) {
> 943           if (abort_fn)
> 944abort_fn(APR_ENOMEM);
> 945  
> 946           return APR_ENOMEM;
> 
> If we returned on line 946, then storage pointed by "pool_allocator"
> will leak.

this is not ganglia, but APR code and seems to not exist anymore on their
SVN version AFAIK, but would assume that APR_ENOMEM means that the
allocation was unsuccesfull anyway and hence probably OK.

to confirm should be better to check with the APR guys instead.

Carlo

PS. removed ganglia-general to avoid crossposting and as it has
    been "handled" in the development version

------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables 
unlimited royalty-free distribution of the report engine 
for externally facing server and web deployment. 
http://p.sf.net/sfu/businessobjects
_______________________________________________
Ganglia-developers mailing list
Ganglia-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ganglia-developers

Reply via email to