Conflicts:
configure.ac
doc/manual.html
runtime/datetime.h
runtime/parser.c
runtime/rsyslog.h
tools/syslogd.c
v4-stable had a bug with RFC5424-formatted structured data, which showed
was detected by the enhanced automatted testbench of v4-beta.
Until now, they were forwarded to processing, but this makes no sense
Also, it looks like the system seems to provide a zero return code
on a UDP recvfrom() from time to time for some internal reasons. These
"receives" are now silently ignored.
A comparison was done between the current and the former source address.
However, this was done on the full sockaddr_storage structure and not
on the host address only. This has now been changed for IPv4 and IPv6.
The end result of this bug could be a higher UDP message loss rate than
necessary (note that UDP message loss can not totally be avoided due
to the UDP spec)
This was a result of an internal processing error if maximum field
sizes had been specified in the property replacer.
Also did some testbench improvements, including omstdout.
... hopefully reducing the number of allocs/frees as well as overall
memory usage in a busy system (plus that these shared properties hopefully
remain in cache longer than its single-instance counterparts...)
This sets stage to enable use of the property-interface to speed
up things (mildly), the next step to be done. I have also fixed one
regression of yesterday's changes.
... plus a fix for a long-time bug in obj-types.h. That lead to
the object pointer only then to become NULL when the object was
actually destructed, I discovered this issue during
introduction of the pRcvFrom property in msg_t, but it potentially had other
effects, too. I am not sure if some experienced instability resulted from this
bug OR if its fix will cause harm to so-far "correctly" running code. The later
may very well be. Thus I will change it only for the current branch and also
the beta, but not in all old builds. Let's see how things evolve.
some things inside the message can be used over a large number of
messages and need to to be allocated and re-written every time. I now
begin to implement this as a "prop_t" object, first use for the inputName.
Some input modules are already converted, some others to go. Will do
a little performance check on the new method before I go further.
Also, this commit has some cleanup and a few bug fixes that prevented
compiliation in debug mode (I overlooked this as I did not compile
for debug, what I normally do, and the automatted test also does not
do that)
... by moving code to stream.c. Thanks to the new design, new cases are
not really needed, resulting in cleaner code.
I also did a cleanup of header file usage as a side-activity.
... and also made it callable via an rsyslog interface rather then
relying on the OS loader (important if we go for using it inside
loadbale modules, which we soon possible will)
The imdiag module now can very effectively inject messages, which also
frees us from uncertainties of tcp reception and processing. All shell
script based tests have been modularized, what makes it far easier to
create new tests. Also, the test bench now executes more reliable and
much faster, because we can now rely on actual engine information where
we previously did just a dumb sleep.
Well, actually this and a lot of related things. I improved the
testbench so that the new capabilities are automatically tested and
also did some general cleanup. The current multiple tcp listener
solution will probably receive some further cleanup, too, but looks
quite OK so far. I also reviewed the way tcpsrv et all work, in
preparation of using this code for imdiag. I need to document the
findings, especially as the code is rather complicated "thanks" to
the combination of plain tcp and gssapi transport modes.
imdiag was never finished (not even really begun), but now I need it.
I made the few things that are available compile, but more serious
work is required.
If the database rejected some entry, making the statement fail on it,
the batch was not cleaned and the same values were retried over and
over, causing a cascade of failures and a denial of service.
We use now OCI_BATCH_ERRORS so that everything valid in the batch is
inserted, and rejected values can be discarded.
Instead of reading a complete line, we'll use a template and delegate
in the core to read such template. Then, all omoracle has to do is to
find that template and use it as the prepared statement.
I'm not sure if this is the correct approach, though. It has to dig
too much into rsyslog's structures...
txt_statement is stored in a private area, so that we don't mess too
much with rsyslog's internals (I still don't feel comfortable with
this much digging into template structures).
This directive controls the amount of memory needed for properties in
the batch. Users should specify the largest value they expect in the
statement. As per Rainer's comment:
on MAX_BUFSIZE: I'd tend to make this configurable, because with
RFC5424 messages can be much longer and RFC5425 now recommends a
minimum maximum size of 8K.
So we let users to choose. Maybe we need a sensible default value to
make users' lifes easier?
Also, the old non-vector based interface is not supported anymore. I
broke it already when moving to this stage.
