closes#86
The uncork events are running from mainloop so these should just
dispatch the right event to the right loop. Doing anything long there
is bad for performance and latency as the uncork list may be huge.
the sync is actually needed when the pages get freed, not precisely
after every item cleanup, as the data technically stays intact until the
deferred free's are called.
This partially reverts commit d617801c31.
The common lockfree doesn't work well for high-volume structures like
eattr cache because it expects the structure to be cleaned up by a
sweeper routine ... which is very ineffective for >1M records.
OTOH, we need the deferred ea_free in all cases ... so keeping that.
There is an IP table for every ROA table, holding special records
combining all known ROAs for every top-prefix.
The ROA digestor is now an IP digestor, running over the auxiliary
table.
The sending channel may be already gone when rte_free_deferred()
is finally called so we have to log about route freeing in the
synchronous call instead.
Channel is now just subscribing to yet another journal announcing
digested tries from the ROA table.
Creating tries in every channel on-the-fly was too slow to handle
and it ate obnoxious amounts of memory. Instead, the tries are
constructed directly in the table and the channels are notified
with the completed tries.
The delayed export-release mechanism is used to keep the tries allocated
until routes get reloaded.
Originally, this mechanism required to check whether there's enough time to work
and then to send an event. This macro combines all the logic and goes more straightforwardly
to the _end_ of the export processing loop.
One should note that there were two cases where the export processing loop
was deferred at the _beginning_, which led to ignoring some routes on
reimports. This wasn't easily noticeable in the tests until the one-task
limit got a ceiling on 300 ms to keep reasonable latency.
In future, this and rtable's data structures should be probably merged
but it isn't a good idea to do now. The used data structure is similar
to rtable -- an array of pointers to linked lists.
Feed is lockless, as with all tables.
Full export (receiving updates) is not supported yet but we don't have
any method how to use it anyway. Gonna implement it later.
Introducing a new omnipotent internal API to just pass route updates
from whatever point wherever we want.
From now on, all the exports should be processed by RT_WALK_EXPORTS
macro, and you can also issue a separate feed-only request to just get a
feed and finish.
The exporters can now also stop and the readers must expect that to
happen and recover. Main tables don't stop, though.
Move bfd_opts grammar inside BFD parser code to avoid dependences between
nest and BFD grammars, which breaks when BFD build is disabled.
Add dummy bfd_opts grammar rule, so protocols can use this nonterminal
even with BFD disabled.
Thanks to Yuri Honegger for the bugreport.
We have now better methods how to measure overall performance
and this obsolete protocol has basically rotten away. If anybody
needs its features, feel free to revive it in future.
To avoid needs for keeping local temporary references for attributes,
now one can use ea_lookup_tmp() to ensure that the attributes are
valid and stored until the task ends. After that, the attributes are
automatically unref'd and also deallocated if needed.
Explicitly marking domains eligible for RCU synchronization. It's then
forbidden to lock these domains in RCU critical section to avoid
possible deadlock.
This commit makes the route chains in the tables atomic. This allows not
only standard exports but also feeds and bulk exports to be processed
without ever locking the table.
Design note: the overall data structures are quite brittle. We're using
RCU read-locks to keep track about readers, and we're indicating ongoing
work on the data structures by prepending a REF_OBSOLETE sentinel node
to make every reader go waiting.
All the operations are intended to stay inside nest/rt-table.c and it
may be even best to further refactor the code to hide the routing table
internal structure inside there. Nobody shall definitely write any
routines manipulating live routes in tables from outside.
In OSPFv3-IPv4 there is no requirement that link-local next hop announced
in Link-LSA must be in interface address range. Therefore, for interfaces
that do not have IPv4 address we can use some loopback IP address and
announce it as a next hop. Also we should accept such address.
Add route attribute gw_mpls_stack to make MPLS stack of route nexthop
accessible from filters. Its type is T_CLIST, which is really not correct
(as it is a list, while T_CLIST is a set). Therefore, we keep this
attribute *undocumented* and it will be *changed* without further notice.
Based on a patch from Trisha Biswas <tbiswas@fastly.com>, thanks!
We can distinguish BGP sessions if at least one side uses a different IP
address. Extend olock mechanism to handle local IP as a part of key, with
optional wildcard, so BGP sessions could local IP in the olock and not
block themselves.
We were, once again, forgetting to mark empty-net journal entries as
processed, as was fixed in 32bb548c11.
Introduced in 548dbb2252.
Caught by cf-ebgp-graceful.
