| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
| |
| |
| | |
Just testing for gcc might not be sufficient
|
| |
| |
| |
| |
| |
| | |
We're calling exit() if the source isn't built with SASL support
if the user tries to use the -S command line argument (causing
some compilers to emit warnings about statement never reached)
|
| | |
|
| | |
|
| | |
|
| |
| |
| |
| |
| | |
Some compilers (at least gcc 4.6.1) may optimize away the code
when using certain optimization levels
|
| |
| |
| |
| |
| |
| |
| | |
If memcached shutdown is requested too early this will lead to
clock_handler not being set up. So actual shutdown will only happen
when some socket event happens which will cause event_hander to be
called which will exit event loop.
|
| | |
|
| | |
|
| |
| |
| |
| |
| | |
It looks like symbol 'interface' is predefined somewhere and makes
compiler confused.
|
| | |
|
| |
| |
| |
| |
| | |
The tap commands should always return error messages just like
the other memcached commands
|
| | |
|
| | |
|
| | |
|
| |
| |
| |
| |
| | |
You might want to run code before and after the engine is started or
stopped. These entry functions allows you to do so.
|
| | |
|
| | |
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
If isasl.pw is changed several times during one second (which is
resolution of mtime in struct stat), then we'll have stale password
database.
To prevent that we keep reloading isasl password for 2 mtime check
cycles. If mtime is still same, we can be sure that we haven't missed
updates.
Change-Id: Ia299aba6296b86ec0e569cf1d13bc9fbede0303a
|
| |
| |
| |
| |
| | |
Uninitialized mutex object will cause the whole memcached process
crashing as soon as it processes stats command.
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Erlang ports don't have API to close only output part. So it seems
that the only reasonable way to reliably control memcached from Erlang
is by signalling shutdown by something else than EOF. EOL seems good
enough. This allows sending shutdown 'message' and waiting for
shutdown termination by observing EOF on memcached's stdout.
Change-Id: I0bc994362e8f55a0956cb0ecb68305e495916ccb
|
| | |
|
| |
| |
| |
| |
| | |
A proxy-engine may want to intercept the calls and should be able
to return an error.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
We are currently using sockaddr_storage to keep track of the address
(to support IPV6), but this caused sendmsg to fail on OpenBSD with
EINVAL. The socket is already in a connected state, so we don't
need to pass in the socket address.
Our test coverage is a bit low for UDP, so I leave the initalization
there for now. I'm pretty sure that we could just set this to zero
here, because we reset the member each time we receive a packet
in try_read_udp().
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
After some idle thought it seems clear enough to just kick the maxconns
deafult down to 1000. It was never 1024 before due to pipes/conns consuming
fds anyway.
Also repairs some tests which were looking for that 1024 number, and some
subtle wording changes to the warnings.
|
| |
| |
| |
| | |
just removed it from Makefile.am
|
| |
| |
| |
| |
| | |
automake's dist target follows symlinks and adds two copies of
libevent into the archive.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
| |
| |
| |
| | |
Kill some outdated ones, make information more accessable in the remaining.
Updated the AUTHORS file to point out that there're more than two.
|
| |
| |
| |
| |
| |
| | |
This is property of memcached and I claim no rights to it.
Signed-off-by: Dustin Sallings <dustin@spy.net>
|
| | |
|
| | |
|
| |
| |
| |
| |
| | |
The client of the interfaces shouldn't have to figure out which system
header to include in order to use our interfaces.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
mcstat pulls statistics from memcached and writes them to stdout.
Recent versions of glibc have a warn_unused_result attribute on fwrite
causing this utility to fail to compile.
While this change in glibc is a bit controversial[0], it's the state
of things on some systems today, so this change explicitly ignores the
result of writes to stdout.
[0]: http://sourceware.org/bugzilla/show_bug.cgi?id=11959
|
| |
| |
| |
| | |
Because people run CentOS.
|
| |
| |
| |
| |
| |
| | |
If setrlimit fails the maximum number of connections is set to
the current filedescriptor limit minus the number of internally
used filedescriptors
|
| | |
|
| |
| |
| |
| |
| |
| | |
Time is relative, memcached time doubly so. set_current_time() sets
the time relative to the start time, which we didn't know until stats
initialization. Mass confusion from doing this wrong.
|