Tom Lane <tgl@sss.pgh.pa.us> wrote:
> For the record, I think this patch is a waste of manpower and we should
> rely on dtrace/systemtap. However, if we are going to make our own
> homegrown substitute for those facilities, a minimum requirement should
> be that it uses the dtrace macros already put into the sources, rather
> than expecting that it gets to clutter the code some more with its own
> set of tracing markers.
How about export dtrace functions as hook function pointers?
For example:
void (*LWLOCK_WAIT_START_hook)(int, int); #define TRACE_POSTGRESQL_LWLOCK_WAIT_START(INT1, INT2) \ if
(LWLOCK_WAIT_START_hook== NULL); else \ LWLOCK_WAIT_START_hook(INT1, INT2) #define
TRACE_POSTGRESQL_LWLOCK_WAIT_START_ENABLED()\ (LWLOCK_WAIT_START_hook != NULL)
If there were such hooks, my profiler could be implemented as
a loadable module on top of the hooks. It might be good to initialize
LWLOCK_WAIT_START_hook with lwlock__wait__start(). If do so, dtrace
probes still work and we can avoid if-null checks for each call.
If acceptable, I'll also suggest new probe functions like
SLEEP, SEND, RECV, SPINLOCK_FAILURE and so on.
Regards,
---
ITAGAKI Takahiro
NTT Open Source Software Center