This change makes the zone table lock-free for reads. Previously, the
zone table used a red-black tree, which is not thread safe, so the hot
read path acquired both the per-view mutex and the per-zonetable
rwlock. (The double locking was to fix to cleanup races on shutdown.)
One visible difference is that zones are not necessarily shut down
promptly: it depends on when the qp-trie garbage collector cleans up
the zone table. The `catz` system test checks several times that zones
have been deleted; the test now checks for zones to be removed from
the server configuration, instead of being fully shut down. The catz
test does not churn through enough zones to trigger a gc, so the zones
are not fully detached until the server exits.
After this change, it is still possible to improve the way we handle
changes to the zone table, for instance, batching changes, or better
compaction heuristics.
The `isc_trampoline` module had a lot of machinery to support stable
thread IDs for use by hazard pointers. But the hazard pointer code
is gone, and the `isc_loop` module now has its own per-loop thread
IDs.
The trampoline machinery seems over-complicated for its remaining
tasks, so move the per-thread initialization into `isc/thread.c`,
and delete the rest.
When the loopmanager is shutting down following a signal,
`dig` and `host` should stop cleanly. Before this commit
they were oblivious to ISC_R_SHUTTINGDOWN.
The `isc_signal` callbacks now report this kind of mistake
with a stack backtrace.
[func] BIND now requires liburcu for lock-free data structures
and concurrent safe memory reclamation. It replaces the
home-grown lock-free linked list and QSBR machinery
added in changes 6108 and 6109. The qp-trie code has
been adjusted to use liburcu.
The first working multi-threaded qp-trie was stuck with an unpleasant
trade-off:
* Use `isc_rwlock`, which has acceptable write performance, but
terrible read scalability because the qp-trie made all accesses
through a single lock.
* Use `liburcu`, which has great read scalability, but terrible
write performance, because I was relying on `rcu_synchronize()`
which is rather slow. And `liburcu` is LGPL.
To get the best of both worlds, we need our own scalable read side,
which we now have with `isc_qsbr`. And we need to modify the write
side so that it is not blocked by readers.
Better write performance requires an async cleanup function like
`call_rcu()`, instead of the blocking `rcu_synchronize()`. (There
is no blocking cleanup in `isc_qsbr`, because I have concluded
that it would be an attractive nuisance.)
Until now, all my multithreading qp-trie designs have been based
around two versions, read-only and mutable. This is too few to
work with asynchronous cleanup. The bare minimum (as in epoch
based reclamation) is three, but it makes more sense to support an
arbitrary number. Doing multi-version support "properly" makes
fewer assumptions about how safe memory reclamation works, and it
makes snapshots and rollbacks simpler.
To avoid making the memory management even more complicated, I
have introduced a new kind of "packed reader node" to anchor the
root of a version of the trie. This is simpler because it re-uses
the existing chunk lifetime logic - see the discussion under
"packed reader nodes" in `qp_p.h`.
I have also made the chunk lifetime logic simpler. The idea of a
"generation" is gone; instead, chunks are either mutable or
immutable. And the QSBR phase number is used to indicate when a
chunk can be reclaimed.
Instead of the `shared_base` flag (which was basically a one-bit
reference count, with a two version limit) the base array now has a
refcount, which replaces the confusing ad-hoc lifetime logic with
something more familiar and systematic.
Ondřej Surý
Michał Kępień
Aram Sargsyan
Petr Špaček
Tony Finch
Mark AndrewsandPetr Špaček
Matthijs Mekking
Evan Hunt
Tom Krizek