In some cases, the logrotate is not fast enough rotating the
logfile after sighub has been send to inputfilegen process. It appears
to happen more often on github CI machines, when it happens,
inputfilegen reopens the NOT yet rotated file and keeps writing
into it. However most messages written until the next sighub are lost
after the logfile finally is rotated.
In order to avoid this, I have added an extra sleep between file
CLOSE and OPEN after sighub with a default of 5ms which appears to work
very well. This should give logrotate enough time to perform the
actuzal rotate.
Also added another parameter to inputfilegen to keep the process
open after finish. Which helps syncing with the testcases.
closes: https://github.com/rsyslog/rsyslog/issues/5000
On NetBSD, time_t has for a long time now been __int64_t.
On 32-bit CPUs, the compiler is not obliged to define
__sync_bool_compare_and_swap_8, so instead this ends up
as an undefined symbol when linking rsyslog. This makes
the code fall back to the pthread / locking method on these
systems, but at least lets the program build.
Due the patch in PR https://github.com/rsyslog/rsyslog/pull/4895
state files are deleted with a 5 second delay in order to fix
missing or duplicated messages. However in INOTIFY mode, we need
an INOTIFY event to trigger a poll_tree that triggers the
delayed deletion. The testcase imfile-statefile-delete.sh will now
create empty dummy files after 6 seconds delay in order to trigger
INOTIFY events.
This fixes & closes https://github.com/rsyslog/rsyslog/issues/4958
rsyslog.conf may affect the host's local name. These changes were
so far only activated after the first HUP. This patch now ensures
that the configured local host name is applied correctly throughout
all processing, including early startup.
This patch causes a slight change of behaviour. However, the behaviour
was inconsitent before. Now it is consistent and according to the config.
Please note: this patch also exposes a global entry point via "regular"
dynamic loading as this makes things much easier to do. This is in-line
with ongoing simplification effort.
Finally, we also remove a CI test that we do no longer need because
the problem covered is now addressed differently and the original issue
can no longer occur.
closes https://github.com/rsyslog/rsyslog/issues/4975
- These tests will help find race conditions hopefully
- fix diag.sh issues running second instance in valgrind mode only
- Add check for minimum valgrind version for new relp tests
Add this both as module an input parameter. Complements already-existing
config param notifyonconnectionclose and mirrors the similar feature from
imptcp.
The module parameter acts as default, similarly to notifyonconnectionclose.
Note that in contrast to imptcp, we emit IP addresses and not host
names. This sticks with the traditional semantics of imtcp.
Note that we also fixed a mislading error message in the case when a
disallowed sender tried to connect.
Thanks to John Chivian for suggesting the addition.
Commit 829a6f6 "CI: improve use of github action matrix feature;
optimization" has introduced use of #pragma GCC diagnostic in function
doNameLine() of runtime/conf.c. This is not supported by old compilers
like GCC 4.4.7 on el6.
During SetString reassign to pThis->szVal.psz, pThis->szVal.psz might not null. It resulted in memory leak and this patch fixes this behaviour.
The problem is mentioned here:
https://github.com/rsyslog/rsyslog/issues/4961
similiar to "abortONUncleanConfig", this parameter aborts rsyslog
when a queue has problems during startup. Some users perfer rsyslog
to terminate in this case. By default, nothing changes.
closes https://github.com/rsyslog/rsyslog/issues/4902
Async service send timeout is not configurable and request cache size is too
small to handle large amount of signing requests with small amount of permitted
requests per aggregation round. For example user with max_requests = 4 results
cache size 5 * max_requests or at least 256. When signing 300 log files cache
will be too small resulting several unsigned blocks. When signing 200 log file
cache will be adequate, but with rate of 4 signatures per second, it is only
possible to sign 4 * 10 blocks before all requests that are not sent out will
timeout.
Fix for the issue is to make send timeout configurable and make the size of the
cache depend on the value of send timeout. New configuration value
sig.block.signtimeout="time, s" introduced that defines the time window wherein
the block has to be signed. The size of the request cache is increased to
3 * max_requests * sign_timeout or at least 256.
There is a worker queue where rsyslog KSI module collects events and signing
requests. When queue is processed thread is periodically put to sleep. Previous
implementation handles signature requests well but sleeps every time after
handling new file open / close event. When several log files are opened or
closed simultaneously process is significantly slowed down. Another issue is
that thread always sleeps 1000ms that may be 2x longer than aggregation round.
This slows down overall signing process.
Fix for the issue is to simply not sleep after file open / close event if there
are next items to be processed. To speed up the signing process, rsyslog uses
KSI aggregator conf. to obtain the aggregation period that is used for the sleep
time configuration.
When a to-be-monitored file is being rotated, some messages may be lost or
duplicated. In case of duplication, many file lines may be duplicated
depending on actual timing. The whole bug was primarily timing depenedent
in general. It most often was visible in practice when the monitored
file was very frequently rotated (we had some report with every few
seconds).
Note that while we try hard to not lose any messages, input file
rotation always has some loss potential. This is inevitable if
the monitored file is being truncated.
Also note that this bugfix affects imfile, only. It has nothing to do
and no relation to rsyslog output files being rotated on HUP.
closes: https://github.com/rsyslog/rsyslog/issues/4797
