in vtim.c

Thilo Keber (thx Thilo)

VSLQ_Flush() vtx_force()s all incomplete transactions to be ready and calls
vslq_process_ready().

vslq_process_ready() retires vtxes until the callback returns with non zero,
so if our dispatch callback does return with an error, flushing will stop,
so we may end up with a non-empty ready list in vslq after calling flush
just once.

To ensure that we flushed everything and vtx_retire()d all vtxes
(which may contain refs to chunk_t_bufs living in malloc()ed memory),
we need to keep on flushing until we know the bowl is empty.

for missing timestamps

the VSL read rate

before calling VSL_Match()

API that use it to prune the tx tree)

filters per record

example if no data were found in the tx, or on buffer exhaustion, etc.

have been causing more work for the reader thread rather than less

enable-debugging-symbols off

the MQ interface, bumps the MQ CURRENT version to 5

thread function), and add some integrity checks

just like the -L and -T options for the standard VSL tools

different in different runs (worker thread returning data to free
lists)

we check for the rdkafka and zookeeper libs (otherwise binaries
depend on them that don't need them)

separately, since they may appear in the log in different orders