hcq/MaxScale
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

MXS-2109: Change semantics of query_classifier_cache_size

Browse Source 
The cache size now refers to the total memory used by the cache instead of
the per thread limit. This makes it easier to use as well as more
predictable by removing the dependency on the number of worker threads.
This commit is contained in:
Markus Mäkelä
2018-10-29 12:01:30 +02:00
parent 06d1189e94
commit 8d1b26060a
3 changed files with 21 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
server/core/config.cc
View File

@ -1209,27 +1209,17 @@ bool config_load_global(const char* filename)
{
log_config_error(filename, rval);
}
else if (gateway.qc_cache_properties.max_size == -1)
{
gateway.qc_cache_properties.max_size = 0;
MXS_WARNING("Failed to automatically detect available system memory: disabling the query classifier "
"cache. To enable it, add '%s' to the configuration file.",
CN_QUERY_CLASSIFIER_CACHE_SIZE);
}
else
{
// Do some post-processing for auto-sized default variables
if (gateway.qc_cache_properties.max_size == -1)
{
int64_t mem_per_thr = get_total_memory() * 0.4 / gateway.n_threads;
mxb_assert(mem_per_thr >= 0);
gateway.qc_cache_properties.max_size = mem_per_thr;
if (mem_per_thr == 0)
{
MXS_WARNING("Could not auto-detect total system memory for the query classifier "
"cache. Manually define `%s` to enable it.",
CN_QUERY_CLASSIFIER_CACHE_SIZE);
}
else
{
MXS_NOTICE("Using up to %s of memory for query classifier cache",
mxb::to_binary_size(mem_per_thr * gateway.n_threads).c_str());
}
}
MXS_NOTICE("Using up to %s of memory for query classifier cache",
mxb::to_binary_size(gateway.qc_cache_properties.max_size).c_str());
}
return rval == 0;
@ -2801,8 +2791,13 @@ void config_set_global_defaults()
gateway.peer_password[0] = '\0';
gateway.log_target = MXB_LOG_TARGET_DEFAULT;
// Note: This is not a valid cache value: it is used to detect that the default value is used
gateway.qc_cache_properties.max_size = -1;
gateway.qc_cache_properties.max_size = get_total_memory() * 0.4;
if (gateway.qc_cache_properties.max_size == 0)
{
// Set to -1 so that we know the auto-sizing failed.
gateway.qc_cache_properties.max_size = -1;
}
gateway.thread_stack_size = 0;
gateway.writeq_high_water = 0;