jrandom authored 20 years ago and zzz committed 20 years ago

this helps avoid catastrophic failures, at least a little, since a failure doesnt kill two sets of tunnels

jrandom authored 20 years ago and zzz committed 20 years ago

for now, lets disable the tunnel pool persistance.  this means that after a router crashes, tunnels it was participating in will fail even if the router comes back up before they expire.
disabling this saves us some IO contention (though this may only be relevent on my kaffe box... dunno)

jrandom authored 20 years ago and zzz committed 20 years ago

always use the cached host/port rather than grabbing the socket's InetAddress (in case it disconnects and throws NPEs)
use the NativeBigInteger as part of the session key negotiation (oops, forgot this one last time)
logging

jrandom authored 20 years ago and zzz committed 20 years ago

lots of bugfixes.  still no rate limiting, but, uh, lots of bugfixes
(release will be packaged and deployed later today)

jrandom authored 20 years ago and zzz committed 20 years ago

workaround an aggressively up-to-spec kaffe implementation (the spec says Socket.getInetAddress() is null if not connected,
but sun lets the getInetAddress() return a value if it had connected then disconnected, while kaffe buggers off and NPEs)

jrandom authored 20 years ago and zzz committed 20 years ago

more tests with the real TCP transport, not just the VM comm system (and for larger sims, dont keepHistory)

jrandom authored 20 years ago and zzz committed 20 years ago

yet another deployment option - the user can define a jbigi.ref environmental variable to specify a file from which the name of the resource to be loaded should be found (default is "jbigi.cfg")
if that file exists, the NativeBigInteger will act as if jbigi.impl was set to the contents of that file.
For instance, a jbigi.cfg containing "win-p4" would have the NativeBigInteger search the classpath for the "win-p4" file and use it as a native library.
The jbigi.ref preempts the jbigi.impl property (only if the file exists and is not empty), but the external platform specific jbigi
preempts this (e.g. jbigi.dll or libjbigi.so), as does the jbigi.enable flag.
This option lets us have the admin console write to a file to choose which jbigi to use, rather than have to parse some shell script, etc

jrandom authored 20 years ago and zzz committed 20 years ago

we *do* still penalize the peer based on tunnel failures, but thats in the reliability calculator, not this one.

jrandom authored 20 years ago and zzz committed 20 years ago

make the tcp connection handler nonblocking by adding another (very short lived) thread - this prevents a peer connecting to us that is very slow (or unconnectable) from forcing other cons to timeout
completely ripped out the fscking bandwidth limiter until i get it more reliable
gave threads more explicit names (for the sim)
logging

jrandom authored 20 years ago and zzz committed 20 years ago

dont synchronize on statics, instead use a seperate format object per context (so large sims dont get bogged down on synchronization)

jrandom authored 20 years ago and zzz committed 20 years ago

add some minimal security to the admin console, requiring a passphrase to be entered when updating the clock offset
this works by a simple substring match of the URL - if the router.config contains the adminTimePassphrase=blah, the time update will only succeed if the URL contains "blah" in it
if the router.config does NOT contain an adminTimePassphrase, the time update WILL BE REFUSED.
aka to use the timestamper, you MUST set adminTimePassphrase AND update the clientApp.0.args= line to include the passphrase in the URL!
e.g.
 clientApp.0.args=http://localhost:7655/setTime?blah pool.ntp.org pool.ntp.org pool.ntp.org

jrandom authored 20 years ago and zzz committed 20 years ago

(yeah, this is what broke cvs HEAD, causing transmission failures, disconnects, encryption errors, etc.  oops)

jrandom authored 20 years ago and zzz committed 20 years ago

logging mods
i really need to rewrite the tcp transport - the code is all functional, but the design sucks.
with the FIFO bandwidth limiter we could get away with a single 'send' thread rather than each TCPConnection having its own writer thread (but we'd still need the per-con reader thread, at least until nio is solid enough)
but maybe the rewrite can hold off until the AMOC implementation.  we'll see