| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
| |
cache constructors/destructors were never used, which just ended up
being wasted branches. Since there's no constructor/destructor for cache
objects, we can use the memory itself for the freelist.
This removes a doubling realloc for the freelist of cache objects and
simplifies the code a bunch.
|
|
|
|
|
| |
FreeBSD version as there seems to be portability issues if you just use
sys/queue, plus some extra features I like in this one.
|
|
|
|
|
|
|
|
|
|
| |
Adding a new type of log entry required updating code in a bunch of
places. Now we use callbacks in the entry structure instead, cutting
down the code and some awkwardness. Should also make it easier for
plugins/extra systems to add log entries.
Moving the writer/parsers to their own section makes the actual logging
code easier to follow, as an added bonus.
|
|
|
|
|
|
|
|
|
| |
Inspired by #768 - logger thread would wake up once per second looking
for watchers or log entries, but without any watchers. Now sleeps on a
condition.
Note that if I ever actually get logger to syslog/stdout/stderr/etc
working properly this will still always poll.
|
|
|
|
|
|
|
|
| |
Add support for `watch connevents` to report opened (`conn_new`)
and closed (`conn_close`) client connections. Event log lines indicate
the connection's remote IP, remote port, and transport type.
`conn_close` events additionally supply a reason for the closing the
connection.
|
|
|
|
|
|
| |
This adds a new field `size` to logger entry lines for item_get,
item_store, and eviction events indicating the size of the associated
item in bytes.
|
|
|
|
|
| |
Note: Do not fix typos in crc32.c because it's copied from an upstream
source
|
|
|
|
|
|
|
|
|
|
|
| |
Fixes #805
Allocate an extra byte for reading the last entry when there is no `\n` at
the end of the file.
Also, check if the user contains null bytes when reading the last entry.
Unrelatedly, add handling in case the auth file size changes while it is being read.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When using NDEBUG assert macro is ineffective which is caught by latest
clang and reports that total_written is set but unused. Therefore check
for NDEBUG to make sure assert is used only when its effective
Fixes
error: variable 'total_written' set but not used [-Werror,-Wunused-but-set-variable]
size_t size = 0, written = 0, total_written = 0;
^
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
|
|
|
| |
my clang vim plugin whatevers creates a .cache directory.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
These changes aim to improve our memcached devtools images to allow us
to reuse them during development stages.
These changes create a memcached user for all the images, switch to this
user by default once memcached have been configured for building.
That allow to run the memcached server directly when the containers are
started, else, without the user the run will fail as we try to execute
memcached as root user.
Also we should notice that the alpine image has been fixed by adding git
to the list of the requirements to install. Else the build will fails.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I had the response code as "HD" in the past, but standardized on OK
while merging a number of "OK-like" rescodes together. This was a
mistake; as many "generic" memcached response codes use "OK". Most of
these are management or specialized uncommon commands.
With this, a client response parser can know for sure if a response is
to a meta command, or some other command.
`-o meta_response_old` starttime option has been added, valid for the
next 3 months, which switches the response code back from HD to OK. In
case any existing users depended on this and need time to migrate.
|
|
|
|
|
|
|
| |
payload length is now part of the main command rather than a flag/token
pair. This allows byte swallow to be more reliable in the case of a flag
parsing error, and makes sets easier to parse and buffer for servers
implementing the protocol.
|
|
|
|
|
|
|
| |
allows returning cas value of the new item on set.
also fixes an extra space that was tagged onto the response line. which
I had not noticed because all of the tests were looking for a space
after the response line :( Bad copypasta from the metaget tests.
|
|
|
|
|
|
|
|
|
|
|
|
| |
needs more testing. in a "pure miss" test to help show hashing overhead,
with four worker threads and large ascii multigets there is a measurable
performance boost for longer keys. Very short keys (10-15 bytes) are
about even.
for 150ish byte keys 12m rps -> 13m rps roughly. Not too bad!
need to test hash table distribution for large sets of keys before
merging. need to consider if worth switching the defaults now or later.
|
|
|
|
|
| |
patch by @sergey-dryabzhinsky on github, but he deleted the repo so I
rewrote the commit.
|
| |
|
|
|
|
| |
if extstore wasn't enabled, crashes. Reported by @zer0e on github.
|
|
|
|
|
| |
This puts syscalls into the allow list that were triggered on latest
systems including extended usage tests, signal and systemd handling.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
incorrectly.
UDP_MAX_PAYLOAD_SIZE actually contains the length of the private UDP header, but resp->tosend only contains the length of the data part.
The number of required UDP packets calculated by the original code will be less than the actual need.
E.g:
1000000/1400 = 714.2 ceil 715
1000000/1392 = 718.3 ceil 719
Actually 719 datagrams are needed, and 715 is wrong.
Signed-off-by: AK Deng <ttttabcd@protonmail.com>
|
| |
|
|
|
|
|
| |
cachedump was the only place in the codebase I can find which copied
the key verbatim. wonder when I can finally remove the command :)
|
| |
|
|
|
|
|
| |
ie: ms [key] b
if 'k' flag is given and key is binary, returns as binary encoded.
|
|
|
| |
small improvements to readme
|
|
|
|
| |
Another issue found by static code analysers
|
| |
|
| |
|
|
|
|
|
|
|
| |
conn_nread state handles c->item like an item, but allow it to be a
temporary malloced blob via setting c->item_malloced = true.
to be used for buffering value reads in the proxy code.
|
|
|
|
|
|
|
| |
new tests were flaky for slow systems; for some reason writing to the
closet socket would exit prove instead of throw any kind of error.
Instead we check that the child pid actually exits and loop a bit to
avoid the race.
|
|
|
|
| |
Detail in: https://github.com/memcached/memcached/issues/741
|
| |
|
| |
|
|
|
|
|
|
|
| |
specifying 'hash' instead of 'all' will make the LRU crawler iterate
over ever bucket in the hash table once, instead of what's in the LRU.
This also doesn't suffer from missing items because of LRU reordering
or high lock contention.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
the idle thread would sleep for up to the max timeout given before
checking for active connections again. the new shutdown code waits on
threads to be joined after a shutdown signal, which could end up
hanging by the idle_timeout.
this now instead uses a cond_timedwait call to allow a condition signal
to wake up the thread to exit.
this should be done for a few other threads that sleep without
conditions (lru_maintainer?)
|
|
|
|
|
|
|
|
|
| |
if you accidentally start memcached with the same options twice,
extstore is initiated before the listener sockets and will happily
truncate its own file.
So this avoids that. Keep in mind any other process can still wipe the
file clean!
|
|
|
|
|
|
|
|
|
|
| |
in the deferred IO patch I'd cribbed some code from another PR which
turned out to use a non-portable return state from getsubopt. On BSD
systems the subopts_value string is simply empty on a non-match.
Fixing this portably looks gross but I couldn't come up with a better
option... and frankly if we're going to pass options through DSO
submodules in the future this is better anyway.
|
|
|
|
| |
was defined in memcached.c, but also used in thread.c.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
By default memcached assigns connections to worker threads in
a round-robin manner. This patch introduces an option to select
a worker thread based on the incoming connection's NAPI ID if
SO_INCOMING_NAPI_ID socket option is supported by the OS.
This allows a memcached worker thread to be associated with a
NIC HW receive queue and service all the connection requests
received on a specific RX queue. This mapping between a memcached
thread and a HW NIC queue streamlines the flow of data from the
NIC to the application. In addition, an optimal path with reduced
context switches is possible, if epoll based busy polling
(sysctl -w net.core.busy_poll = <non-zero value>) is also enabled.
This feature is enabled via a new command line parameter -N <num>
or "--napi_ids=<num>", where <num> is the number of available/assigned
NIC hardware RX queues through which the connections can be received.
The number of napi_ids specified cannot be greater than the number
of worker threads specified using -t/--threads option.
If the option is not specified, or the conditions not met, the code
defaults to round robin thread selection.
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
|
|
|
|
|
|
|
| |
libevent 1 doesn't have the socket id wrapper. since version 2 is 10
years old we should fail to build on version 1.
explicit cast to avoid possible pedantic build flags complains (ie can be the old iovec interface)
|
|
|
|
|
|
| |
since multiple queues can be sent to different sidethreads, we need a
new mechanism for knowing when to return everything. In the common case
only one queue will be active, so adding a mutex would be excessive.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
mc_resp is the proper owner of a pending IO once it's been initialized;
release it during resp_finish(). Also adds a completion callback which
runs on the submitted stack after returning to the worker thread but
before the response is transmitted.
allows re-queueing for pending IO if processing a response generates
another pending IO. also allows a further refactor to run more extstore
code on the worker thread instead of the IO threads.
uses proper conn_io_queue state to describe connections waiting for
pending IO's.
|
|
|
|
|
|
| |
reserve space in an io_pending_t. users cast it to a more specific
structure, avoiding extra allocations for local data. In this case what
might require 3 allocs stays as just 1.
|
|
|
|
|
| |
don't gate on EXTSTORE for the deferred io_queue code. removes a number of
ifdef's and allows more clean usage of the interface.
|
|
|
|
|
|
|
| |
extstore.h is now only used from storage.c. starting a path towards
getting the storage interface to be more generalized.
should be no functional changes.
|
|
|
|
|
|
|
|
|
|
|
|
| |
want to reuse the deferred IO system for extstore for something else.
Should allow evolving into a more plugin-centric system.
step one of three(?) - replace in place and tests pass with extstore
enabled.
step two should move more extstore code into storage.c
step three should build the IO queue code without ifdef gating.
|
|
|
|
| |
Also changes the maxconns_fast test to check against the highest fd
|
|
|
|
|
|
|
|
|
| |
of course the previous commit didn't actually fix the bug: during
poll() call we weren't filtering for the right watcher when we were
polling a specific one. This could lead to reading a POLLHUP on the
wrong socket and other issues.
hopefully that solves some mysteries with the watcher system.
|