tag name | 1.3.3 (db2024922d5c979a449f63c45df6895ca1451192) |
tag date | 2009-04-03 17:07:32 -0700 |
tagged by | Dustin Sallings <dustin@spy.net> |
tagged object | commit b53bf8b576... |
download | memcached-1.3.3.tar.gz |
---|
Release 1.3.3 (a beta)
1 Download
~~~~~~~~~~~
Download Link:
http://memcached.googlecode.com/files/memcached-1.3.3.tar.gz
2 Features
~~~~~~~~~~~
2.1 Can set listen backlog on the commandline.
===============================================
Prevent connection refused during connection storms at the cost of
kernel memory.
2.2 stats settings
===================
Show all current server settings (useful for troubleshooting as well
as internal verification).
3 Bug fixes
~~~~~~~~~~~~
- Alignment bug in binary stats (bug26)
- Occasional buffer overflow in stats (bug27)
- Try to recycle memory more aggressively. (bug14)
- incr validation (bug31)
- 64-bit incr/decr delta fixes (bug21)
- ascii UDP set (bug36)
- stats slabs' used chunks (bug29)
- stats reset should reset item stats, eviction counters, etc... (bug22)
- Fix all stat buffer management
4 Misc
~~~~~~~
- More tests
- More/better documentation
- Code cleanup
5 Stable fixes from Dormando
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
5.1 New Stats
==============
5.1.1 accepting_conns
----------------------
1 or 0 to indicate whether the server is currently accepting
connections or not.
The server will stop accepting connections when it has as many as it's
configured to take.
5.1.2 listen_disabled_num
--------------------------
The number of times socket listeners were disabled due to hitting the
connection limit.
5.1.3 cmd_flush
----------------
The number of times the flush command was issued.
5.2 missing key debugging
==========================
With verbosity enabled, you can see *why* objects were not found. In
many cases, an item exists under a given key, but is considered
invalid due to lazy expiration or flush.
5.3 tail repair
================
There is a rare, unidentified reference leak that causes a slab to be
full of invalid objects that cannot be evicted via the LRU nor will
they expire on their own.
Tail repair is a strategy by which we forcefully evict objects that
are marked as ``in-use'' (that is, in-flight or otherwise being used),
but haven't been accessed in a long time (currently three hours).
There is an additional stat that comes along with this (tailrepairs on
a slab) that will allow you to detect that this condition has occurred
on one of your slabs.
5.4 socket listen bugs
=======================
There were some issues listening to sockets on machines with different
network interface configurations (i.e. no network, only ipv4, only
ipv6, etc...).
6 Contributors
~~~~~~~~~~~~~~~
The following people contributed to this release since 1.3.2. Please
refer to the 1.3.2 release notes for more info:
http://code.google.com/p/memcached/wiki/ReleaseNotes132
28 Dustin Sallings
8 Trond Norbye
6 dormando
5 Brad Fitzpatrick
4 Steve Yen
1 Eric Lambert
1 Clinton Webb
1 Chris Goffinet
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Darwin)
iEYEABECAAYFAknWpPgACgkQeWDnv0PlnVR0FwCdEdmloUzoQaxtFG6xBaiFfZd1
7DsAniCj46zypeg+8Q7xr8CrCmvvloSd
=yMx3
-----END PGP SIGNATURE-----