jbuchbinder
commented
12 years ago From original issue:
In gmond.c, function Ganglia_metadata_save(...) and Ganglia_value_save(...):
metric and metric->name memory is allocated from host->pool.
When gmetric data timed out and is deleted in Cleanup_data(...), metric is
removed from gmetric hash table and metric->pool is destroyed. But allocated
memory for metric and metric->name can't be released unless host->pool is
destroyed. If gmond receive same gmetric data again, metric and metric->name
alway allocate new memory. If certain gmetric data switchs from time out/active
frequently, memory usage can increase more than 40M per day.
My temporary solution is to comment out following code in Cleanup_data to avoid
allocating memory again:
//apr_hash_set( host->metrics, metric->name, APR_HASH_KEY_STRING, NULL);
//apr_hash_set( host->gmetrics, metric->name, APR_HASH_KEY_STRING, NULL);
/* destroy any memory that was allocated for this gmetric */
//apr_pool_destroy( metric->pool );
I think metric->name can be allocated from metric->pool, but metric may need a
new pool to hold.
Thanks
Wei
vvuksan
keyurdg
denji
As discussed earlier on the ganglia-developers list
Please give this a bump in priority.
Our gmond steadily increases to multiple Gigabytes memory usage