Currently Scribunto creates profiling data on page parses, that is included in the html source code (see below for an excerpt). Working on the performance of the Wikibase Lua components, it would be helpful if we collected (in Graphite, so that we can create a nice Grafana dashboard) this data.
In more concrete terms, it would be nice to have a whitelist of functions to track, for which we will track how much time is spent in them globally (and maybe how often they're called). On top of that, we should collect the overall runtime of all Lua code (so that we have a reference metric, to set the per-function data in context). With that we could adequately measure the impact of Scribunto related changes.
Lua time usage: 6.217/10.000 seconds
Lua memory usage: 11.68 MB/50 MB
Lua Profile:
Scribunto_LuaSandboxCallback::callParserFunction 2840 ms 46.1%
Scribunto_LuaSandboxCallback::getEntity 900 ms 14.6%
Scribunto_LuaSandboxCallback::addStatementUsage 700 ms 11.4%
Scribunto_LuaSandboxCallback::getEntityStatements 320 ms 5.2%
Scribunto_LuaSandboxCallback::gsub 220 ms 3.6%
init <Modul:Languages> 120 ms 1.9%
recursiveClone <mwInit.lua:41> 120 ms 1.9%
Scribunto_LuaSandboxCallback::getAllExpandedArguments 100 ms 1.6%
Scribunto_LuaSandboxCallback::incrementStatsKey 80 ms 1.3%
getExpandedArgument <mw.lua:165> 60 ms 1.0%
[others]