--- AnyEvent-MP/Changes 2009/09/08 13:46:25 1.60 +++ AnyEvent-MP/Changes 2012/03/01 19:37:59 1.121 @@ -1,19 +1,150 @@ Revision history for AnyEvent::MP -TODO: global protocol fix? -TODO: use seednodes only to broadcast? TODO: testsuite TODO: intro: maybe simple job pool example? TODO: intro: mention watchdog... -TODO: group monitoring +TODO: maybe disbale warnings by default? +TODO: listener-scopes (10.0.0.1:4040@vpn) and connect-scopes ("vpn,public") -TODO: cal, cal_func -TODO: mon_nodes should init-call for all up nodes -TODO: better autoseeding, timout*(nconnect+1) -TODO: fix the "trying to contatc node under wrogn address" problem. -2009-09-08 03:58:42 <9> trying connect to seed node 10.0.0.18:4040. -2009-09-08 03:58:42 <9> 10.0.0.18:4040 frank already connected, not connecting again. +faq: can't se anything +faq: all is asynch +faq: how to interface to non-perl nodes? + +TODO: update docs +TODO: secure - do not allow remote code execution +TODO: switch to AnyEvent::Log +TODO: Global loaded on demand only, adds grp database and node address database +TODO: transport connects use any global node to find addresses (maybe itself) +TODO: maybe check g_proto version, if sent? how about node protocol? +TODO: connect instantly in node->connect, not with timer-0 +TODO: maybe switch to md5+sha1+sha256 or so... +2 + - INCOMPATIBLE CHANGE: the node protocol used by this version is not + compatible to versions <2 - you have to upgrade all nodes. + - AnyEvent::MP::Global is now optional - only seed nodes autoload it + and create a fully-meshed network with each other, normal nodes only + have to connect to seed nodes, other connections are on-demand. + - new "norc" and "force" options for MP::configure to ignore + or override the rc file. + - move seed code to Kernel. + - fixed small typo in MP::Transport. + - fix (unused) provider-kv (version sometimes missing). + - reduce the default export list of MP::Kernel. + - reduced various random delays to zero, as they + are hopefully no longer needed with the new global + protocol. + - do not use GMP for generating alphanumeric strings, + use base64 with _ and - as extra chars instead. + - anonymous node names and remote port names have been shortened + considerably. + +1.30 Thu Jun 30 11:30:39 CEST 2011 + - connection errors at the right time would kill node + connections in a bad way (patch by Malcolm Studd). + +1.29 Fri May 7 20:13:39 CEST 2010 + - codename "Sadrak". + - fix error in callback at AnyEvent/MP/Global.pm line 339 + (found by Sadrak). + - listener-less nodes had trouble sending keepalive + messages on write timeouts (found by Sadrak). + - the monitor guard could cause a memleak due + to autovivification if a mon was cleared after + the port was gone (analysed by Sadrak). + - do not overwrite the config file if we couldn't read it + for some reason (lesson demonstrated by Sadrak). + +1.28 Thu Apr 1 21:23:54 CEST 2010 + - accepted connections didn't correctly set up the SRCNODE + in some cases, leading to intra-node messages (such as monitoring) + to fail silently (great testcase by Sadrak). + +1.27 Sat Mar 20 21:11:16 CET 2010 + - avoid exceptions in the transport code if the connection + fails during TLS negotiation. + - delay new address broadcasts in seednodes, to reduce the + likelyhood of nodes connecting at the same time, leading to + message loss. + - improve general message passing efficiency by 8%, and special + case json framing for a 50% speed increase. + - add hints how to interface to aemp nodes from other programming + languages using the aemp protocol in AnyEvent::MP::Transport + (it can really be quite simple). + - handle unsupported remote framings more gracefully. + +1.26 Wed Dec 30 16:48:37 CET 2009 + - INCOMPATIBLE CHANGE: the high-level node protocol changed, + all nodes need to upgrade. + - remote kil's did not work (reported by Eugene Grigoriev). + - documented monitor_timeout, connect_interval, auth_offer, + auth_accept, framing_format, autocork and nodelay options + and make them configurable via bin/aemp. + - some doc updates. + - remote kil's were not synchronous to message sends which affected + nobody, as remote kils were broken. + - remote mon/unmonitor requests were not synchronous to message + sends either sometimes, although it likely didnt matter. + +1.24 Wed Dec 9 14:59:52 CET 2009 + - codename "the foul egg". + - increased "unmonitored local port died" message level to 2, so + it is displayed by default now. + - generate random seed at configure time, not module loading time, + so fork&configure works as expected. + - decreased "seed"-related message priority in Global from 0 (d'oh) + to 9, as they are not signifcant. + - new experimental module AnyEvent::MP::DataConn to establish + additional data connections between nodes. + - bin/aemp now loads AnyEvent::Watchdog::Util before trying + to call restart. + - bin/aemp setcert didn't properly handle the filename argument. + - removed empty and unused timeout= parameter from protocol greeting. + +1.23 Fri Nov 6 18:46:26 CET 2009 + - listener-less nodes were misinformed by their masters about + the location of new nodes, and therefore didn't connect + properly in all cases. + - messages send during node-up processing could sometimes get lost, + which would hamper global's ability to mesh the network. + - fixed AnyEvent::MP::Global::grp_reg to return a guard also + in scalar context. + - fixed AnyEvent::MP::Kernel::mon_nodes to return a guard also + in scalar context. + - try to improve error reporting when automatically loading + a function - being unable to load a module will now + stop the process and report the error. + +1.22 Sat Oct 17 03:41:47 CEST 2009 + - fix two rcv-bugs: after adding a tagged rcv, the default + callback would not be used or changed anymore (reported + by Felix Antonius Wilhelm Ostmann, with apologies for + taking so long to releae the fix). + - added Coro::MP introduction to AnyEvent::MP::Intro. + - add peval function. + +1.21 Mon Oct 5 21:58:36 CEST 2009 + - always delay kil messages. + - remote monitor requests were never cleaned up (found by + Sadrak and analysed by elmex). + - move $VERSION to AnyEvent::MP and AnyEvent::MP only. + +1.2 Thu Oct 1 22:59:35 CEST 2009 + - fix a rather big bug where message processing could be + inhibited completely. + - grp_mon could not return a guard due to a typo. + - implement primitive aemp shell. + - do not use bound address as listen length (reported by + Felix Antonius Wilhelm Ostman). + - no longer export an undefined "reg" function from AE::MP. + - the mon_guard return value no longer keeps an additional + reference to the passed refs. + +1.1 Fri Sep 11 04:34:03 CEST 2009 + - bumped the transport protocol to version 1 - sorry, but there + were too many bugfixes/changes. + - new function AnyEvent::MP::cal for simple rpc. - renamed AnyEvent::MP::Global functions to grp_reg|get|mon. + - implemented monitoring for Global groups. - removed all userspace time-outs from the transport, instead rely on tcp retransmit timeouts and tcp keepalive. - spawn now delays spawns on the local node artificially, so @@ -30,6 +161,9 @@ - try to backport to 5.8.2, unfortunately, recursive dependencies rely on Module::Build, which fails at 5.8.7, so YMMV. - use a vastly different reseeding strategy. + - use seed nodes to notify nodes about new nodes. + - in the case where a ndoe learns new addresses after it tried + connecting, prefer the new addrseses over the old ones. 1.0 Wed Sep 2 20:40:43 CEST 2009 - randomise more connection times, but reduce delay to 0.2s (seeds)