Home

Jump to bottom Edit New page

Kiyoshi Ikehara edited this page May 16, 2013 · 44 revisions

EN / JP

Flare

Flare is a distributed, and persistent key-value storage compatible with memcached, with several additional features (as follows):

persistent storage (you can use flare as persistent memcached)
pluggable storage (Tokyo Cabinet and Kyoto Cabinet -experimental- are currently available)
data replication (synchronous or asynchronous)
data partitioning (automatically partitioned according to the number of master servers (transparent for clients)
dynamic reconstruction, and partitioning (you can dynamically (I mean, without any service interruption) add slave servers and partition master servers)
node monitoring and failover (if any server is down, the server is automatically isolated from active servers and another slave server is promoted to master server)
request proxy (you can always get same result regardless of servers you connect to, so you can think of a flare cluster as one big key-value storage)
over 256 bytes keys, and over 1M bytes values are available

Flare is free software base on GNU GENERAL PUBLIC LICENSE Version 2.

News

[2013/04/16] 1.0.17.1 release

released 1.0.17.1

1.0.17.1

feature updates
flarei: sanity check the thread_type value in flare.xml
bug fixes
fixed a duplicate thread id issue

[2013/02/18] 1.0.17 release

released 1.0.17

1.0.17

Feature updates:
added support for Kyoto cabinet
binary protocol implementation and input tests
touch/gat implementation
migration from gethostbyname to getaddrinfo
integrated valgrind memory tests
added over 2000 unit tests
Bug fixes:
various fixes under BSD
append does not reset the stored flag anymore
fixes for input validation

[2013/01/21] 1.0.16.1 release

released 1.0.16.1

1.0.16.1

Bug Fix
fixed a rare race condition which could lead a replication process to end prematurely.

[2012/10/16] 1.0.16 release

released 1.0.16

1.0.16

Feature Updates
add --noreply-window-limit (default=0) for high latency network
Bug Fixes
fix connection::readline() to check the last character of a read buffer
fix thread safety of utility functions
fix reconstruction-bwlimit option

1.0.15

Feature Updates
added --key-hash-algorithm option (default=simple) -- (only applicable for new or single-partition clusters)
Bug Fixes
fix wrong post proxy destination during dynamic patition creation
fix unintentional thread triggering

1.0.14

Feature Updates
added 'stats threads queue' op
added 'shutdown' op (for graceful shutdown)
added --max-total-thread-queue option (default=0 (unlimited))
added --proxy-prior-netmask option
added --storage-dfunit (support for dynamic defragmentation)
Bug Fixes
fix double unlock (which causes dead lock)
fix dump error handling (tc iteration returns error in some rare cases)

1.0.13

Feature Updates
added 'show variables' op
added reconstruction-bwlimit option (node server only, in KB)
Bug Fixes
set default values to keep compatibility w/ older index servers
fix bug of replication inconsistency
fix possible memory leaks
fix possible node map corruption, happens when non-existent nodes are passed to remove_node()
fix segfault (cased by multi get w/ same keys (like "get key1 key1 key1"), w/ concurrent expiration)

[2011/03/01] 1.0.12 release

released 1.0.12

Bug Fixes
fix replication bug (esp.important when reconstructing nodes)
fix error handling in "keys" op

[2010/11/20] 1.0.11 release

released 1.0.11

Feature Updates
added --partition-size option (default=1024)
added --partition-modular-virtual option (default=4096)
Bug Fixes
fix possible dead locks (could happen when activating new master node under heavy load)
suppress bogus master reconstruction, happens when slave is promoted to master