Page MenuHomeClusterLabs Projects
Files F3151581

No OneTemporary
Actions

View Options

diff --git a/TODO b/TODO
index a0b47a96..a68f4af2 100644
--- a/TODO
+++ b/TODO
@@ -1,92 +1,91 @@
debian packaging (mostly resync from rpm packaging):
- (sync) Check policy on /etc/pam.d/* install/usage
- (sync) Make groupadd kronosnetadm silent and dynamic
- - (sync) systemd packaging in debian?
- (sync) install /etc/default/kronosnetd
rpm packaging:
- (issue) check why systemd doesn't let me set scheduler priority
fixed for systemd unit file, LSB init + systemd still fails
https://bugzilla.redhat.com/show_bug.cgi?id=893015
link/host level:
- (issue) implement packet fragmentation via IP_MTU_DISCOVER/IP_MTU, is the only
solution to avoid conflicts with tun/tap mtu setting and various
others buffer sizes.
- (issue) sort out read buf from recv_from_links since it might be too small vs
sending out 128k (+header+encryption+....). this is also related
to max PMTU
- (rfe) compress: should only compress user data, we will add a bit in the data
header to indicate if the pckt is compressed or not (save time).
this approach allow runtime change of compress.
open questions are: methods? level? zlib? lzo? bz? lzma? xz?
how much do we save by compressin our header?
compress must happen before encrypt
we can express compress data in packet type without adding extra
flags to the headers. DATA -> BZ/GZDATA and we can change that
right before encrypting. Using a similar approach to PING_MASK
- (rfe) crypto: expand API to support dual key for rekey process
- (issue) crypto: we need to port again "crypto header size" from corosync
necessary for MTU handling
- (rfe) link id made optional? right now we need the link id to match
on both sides of the connection. this is somewhat annoying
from a user perspective. Evaluate if we can make it optional.
- (rfe) make hostid autogenerated in a consistent way?
- (issue) need bind to interface for dynamic ip local interfaces vs src ip
address or find a way to autodetect the new ip on that interface
(listen to kernel netlink?)
- (issue) must implement link auth via user/passwd. This is necessary
in case key is leaked.
- (rfe) Check IPV6_NEXTHOP for v6 sockets and find equivalent for v4 (Jesper?)
this would allow using one IP address as destination via multiple links
- (rfe) add statistics at different levels (pckt per host/link, bytes, crypto
overhead, frame overhead, pure data...)
- (issue) pong count to consider a link active and stable
(avoid excessive flapping if a link is unstable)
- (rfe) link connection access-list
(chrissie is working on generic code for this one)
- (rfe) improve host-to-host communication. Right now I am not satisfied with
the current implementation, even if it works.
- (rfe) implement link switching via scoring system based on:
1) latency
2) priority (auto/manual)
3) usage (over XX% traffic start RR)
4) flapping of the links (time/sec)
this requires complex rules setting and a super efficent way to look
up destination links
- (rfe) benchmark tests for all critical paths in switching threads
libknet:
- (issue) review locking handling policies (attempt locks vs deadlocks and so on)
almost done, needs knet_link_* bits done
- (issue) add .3 man pages
libtap:
- (rfe) consider adding dhcp support for tap device
it can be done now via up.d/ scripts, but it's not intuitive
- (issue) add .3 man pages
- (issue) improve tests to cover thread safety and better error codes
specially from the up/down handling.
kronostnetd:
- (issue) beside the code that is as bad as it can possibly be and
will make you wish to have a tea spoon handy to carve your
eyeballs out, the vty needs a good clean/rewrite
- (rfe) add logging config (per subsystem/global)
- (issue) missing output from several command execution failures in vty mode
- (issue) fix check_param for ip/prefix/crypto
(this is part of the rewrite as it needs more clever arg parsing
code/method)
- (rfe) split vty_cmd_files to be smaller. it's just too big to handle
nicely.
- (rfe) add equivalent of "description: ...." to various levels
- (issue) fix config file format. current one will make you scream
- (rfe) add optional options. right now it's necessary to specify everything
all the time.
- (rfe) implement tab completion on options
general:
- (rfe) check code with coverity
- (issue) missing unit tests on many many bits
- (issue) missing docs of all kind, devel, users, admin guide.

File Metadata

Mime Type
text/x-diff
Expires
Mon, Feb 24, 7:03 AM (1 d, 5 h ago)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1459967
Default Alt Text
(4 KB)

Event Timeline