[erlang-questions] better profiler for erlang / ways to troubleshoot a running system
Tue Dec 9 21:55:56 CET 2008
There is already support for following a "forlopp" of messages
originating from a certain point.
Take a look at the seq_trace module.
The sequential tracing can be activated by code that you prepare
yourself in your app or
by use of match specifications see
where you se an example of how to activate sequential tracing without
your application code. Just trace on something suitable where you can activate
the sequential trace.
Note it is Erlang messages that creates the sequence going from Erlang
process to Erlang process.
/Kenneth Erlang/OTP, Ericsson
On Tue, Dec 9, 2008 at 7:43 PM, Joel Reymont <joelr1@REDACTED> wrote:
> I had a terrible bout with the Erlang profiler recently and a very
> poor experience trying to understand the bottlenecks and issues while
> scaling OpenPoker.
> I'm wondering, how much of a market is there for a commercial Erlang
> profiler or tool that lets you peer into a running system, trace
> packets end to end, etc.
> For example, I badly wanted to tag a certain packet incoming over a
> socket and then timestamp it as it goes through the system, until a
> reply packet is returned to the socket. I wanted to measure the time
> it took my packet to go from function to function but couldn't.
> I envision loading a trace message dump into triple store, gigabytes
> of it, and then querying the data in various ways to better understand
> the behavior of my system. A triple store holds subject/predicate/
> object triples which map well to source pid/message/destination pid. A
> good triple store should hold gigabytes of data with no trouble.
> erlang-questions mailing list
More information about the erlang-questions