OldHome

flare

rough note for developer(= me)

• check procs around an iteration lock

• index server failover

• shrink partition support

• fix deamonize sequence
  • restart very frequently fails due to index node cannot detect node down → update error messages
  • send node state=down messages in case of graceful shutdown (not yet confirmed)

• connection::write() does not retry in case of EAGAIN w/ specific case (not yet verified)

• queue retry/failover architecture (this should be very important to achieve complete redundncy, robustness and reliability)
  • update logs
  • queue failure logs
  • re-queue in case of node down detection

• threads per request → libevent

• file system implementation w/ FUSE

• on memory storage (like memcached)
• luxio storage

• think about table database
• support other data models (e.g. key:value = 1:n, something like redis)

• more flexible distribution (algorithm based w/ arbitary weight for each node, mapping based…)
• standby for index server

• reliable client library

• performance tunes (get rid of shared_ptr, and bogus memory copies)
• doc comments

archive (anyway)

• monitoring thread (handler_monitor in lib) (thread_type=thread_type_monitor)
  • adding monitor-interval (sec)
  • monitoring thread waits for monitor-interval sec w/ thread::queue_wait() (or something like that) – pthread_cond_timedwait()
  • if timedout, send ping and if not, do somes ops according to queue
• queue framework in thread (pass derived thread_queue object and down cast w/ dynamic_cast)
  • (perhaps) I do not pass op objects directly to thread queue
  • queue object will be handled w/ shared_ptr

• implement “stats nodes”
• implement “stats monitors”(i do not need this, anyway)

• state=(active|standby|…?)
• adding role=(master|slave|proxy…)

• log message clean up (defered:)

• handling node down/up event
• “node role” op
• “node state” op
• “node remove” op
• “set” op
• “get” op
• “delete” op
• “add” op
• “replace” op
• “cas” op
• “incr” op
• “decr” op
• “vervosity” op

• implement proxy framework (w/ proxy-concurrency ini option)

• node reconstruction
• support read only activate of new partition

• fix delete op (a little bit…)

• stress tests (and debug)

• stats update hooks

• 64 bit fixes

• tchoptimize/tchtune support

• mysql replication (experimental)

• documentation (in english) (anyway)

• index server stalls when one or mode node get network-unreachable(like “no route to host”)? – more investigation required

• key dump op

• support uds

• fix dynamic key balancing algorithm

• stats threads causes segfault under heavy load (→ perhaps fixed but not yet verified)

• tcb storage

• management tools
  • slave delay detection

• first proxy request sometimes slow down