routing.c fixed to properly remove rate-limited gossip_store entries
when channels are closed. This caused gossipd to crash on a subsequent
gossip_store_load. Also corrects an overzealous limit of one gossip_store
entry per message (should now allow one broadcastable and one
rate-limited). Addresses issues 5387, 5395.
Changelog-None
routing.c now flags rate-limited gossip as it enters the gossip_store but
makes use of it in updating the routing graph. Flagged gossip is not
rebroadcast to gossip peers.
Changelog-Changed: gossipd: now accepts spam gossip, but squelches it for
peers.
This will be used to decouple internal use of gossip from what is
passed to gossip peers. Updates GOSSIP_STORE_VERION to 10.
Changelog-Changed: gossip_store updated to version 10.
@whitslack complained of large CPU usage by connectd at startup;
I ran perf record on connectd on my machine (which sees a little spike, only)
and I see the cost of reading and discarding the entries:
```
- 95.52% 5.24% lightning_conne lightning_connectd [.] gossip_store_next
- 90.28% gossip_store_next
+ 40.27% tal_alloc_arr_
+ 22.78% tal_free
+ 11.74% crc32c
+ 9.32% fromwire_peektype
+ 4.10% __libc_pread64 (inlined)
1.70% be32_to_cpu
```
Much of this is caused by the search for our own gossip: keeping this separately
would be even better, but this fix is minimal.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Fixed: connectd: reduce initial CPU load when connecting to peers.
And turn "" includes into full-path (which makes it easier to put
config.h first, and finds some cases check-includes.sh missed
previously).
config.h sets _GNU_SOURCE which really needs to be done before any
'#includes': we mainly got away with it with glibc, but other platforms
like Alpine may have stricter requirements.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Before:
Ten builds, laptop -j5, no ccache:
```
real 0m36.686000-38.956000(38.608+/-0.65)s
user 2m32.864000-42.253000(40.7545+/-2.7)s
sys 0m16.618000-18.316000(17.8531+/-0.48)s
```
Ten builds, laptop -j5, ccache (warm):
```
real 0m8.212000-8.577000(8.39989+/-0.13)s
user 0m12.731000-13.212000(12.9751+/-0.17)s
sys 0m3.697000-3.902000(3.83722+/-0.064)s
```
After:
Ten builds, laptop -j5, no ccache: 8% faster
```
real 0m33.802000-35.773000(35.468+/-0.54)s
user 2m19.073000-27.754000(26.2542+/-2.3)s
sys 0m15.784000-17.173000(16.7165+/-0.37)s
```
Ten builds, laptop -j5, ccache (warm): 1% faster
```
real 0m8.200000-8.485000(8.30138+/-0.097)s
user 0m12.485000-13.100000(12.7344+/-0.19)s
sys 0m3.702000-3.889000(3.78787+/-0.056)s
```
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This is better than using the previous "keep statting the file" approach,
since we can also tell you how long the replacement is, to avoid a
gratitous load.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
They're not defined to be, though we've not seen this on Linux (testing
showed that it is page-level atomic, which means it can still happen across
page boundaries though!). This was pointed out by whitslack in
https://github.com/ElementsProject/lightning/issues/4288
In practice, this just means not complaining when it happens, and also
not trying to get tricky to use it on MacOS (we can safely seek & write,
since we're single-threaded).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Removed: Removed bogus UNUSUAL log about gossip_store 'short test'.
If gossip_store is an incorrect version, we will recreate it: with an incorrect
version! This means we never get persistent gossip, *and* the pay plugin will
fail to map the gossip_store. Everyone will be sad.
Debugged-by: Matt Whitlock
Fixes: #4376Fixes: #4288
Typing-by: Rusty Russell <rusty@rustcorp.com.au>
This overcomes the internal spam filter on updates, which can be useful
if we're actually trying to send through such a node.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Fixed: Protocol: always accept channel_updates from errors, even they'd otherwise be rejected as spam.
Fixes: #4300
Instead of a boutique message, use a "real" channel_announcement for
private channels (with fake sigs and pubkeys). This makes it far
easier for gossmap to handle local channels.
Backwards compatible update, since we update old stores.
We also fix devtools/dump-gossipstore to know about the tombstone markers.
Since we increment our channel_announce count for local channels now,
the stats in the tests changed too.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This avoids overwriting the ones in git, and generally makes things neater.
We have convenience headers wire/peer_wire.h and wire/onion_wire.h to
avoid most #ifdefs: simply include those.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This is a better fix than doing it manually, which turned out
to do it in the wrong order (node_announcement followed by
channel_announcement) anyway.
Should fix many "Bad gossip" messages.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
It usually means we're missing something, but there's no way to ask what.
Simply start a broad scid probe.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Since we have to validate, there can be a delay (and peer might
vanish) between receiving the gossip and actually confirming it, hence
the use of softref.
We will use this information to check that the peers are making progress
as we start asking them for specific information.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This is the modified-time of the file. We have to store it internally
since we overwrite the gossip file with compaction on startup.
This means the "are we behind on gossip?" heuristic is no longer inside
gossip_store.c, which is cleaner.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We've been slack, but it's going to be important for testing
ratelimiting. And it currently has a minor memory leak.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We ignored this before, which meant that the DEVELOPER-mode check that we
delete the correct record didn't check that it wasn't already deleted.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We always know the length, so we don't need it. It causes much extra work
when we want to delete a record, which I suspect may cause issues amongst
some users who've been seeing gossip_store corruption.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Rewriting the gossip_store is much more trivial when we don't have
any pointers into it, so add some simple offline compaction code
and disable the automatic compaction code.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
The crashes in #2750 are mostly caused by us trying to partially truncate
the store. The simplest fix for release is to discard the whole thing if
we detect a problem.
This is a workaround: it'd be far nicer to try to recover.
Fixes: #2750
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
If something went wrong and there was an old one, we were
appending to it!
Reported-by: @SimonVrouwe
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We might have channel_announcements which have no channel_update: normally
these don't get written into the store until there is one, but if the
store was truncated it can happen. We then get upset on compaction, since
we don't have an in-memory representation of the channel_announcement.
Similarly, we leave the node_announcement pending until after that
channel_announcement, leading to a similar case.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We can't continue, since we've moved the indexes. We'll just crash
anyway, as seen from bugs #2742 and #2743.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
I decided to try a faster implementation, only to find our crc32c was
not correct! Ouch.
I removed the crc32c functions from ccan/crc, and added a new crc32c
module which has the Mark Adler x86-64-optimized variants.
We bump gossip_store version again, since csums have changed.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
(Or, if we crashed before we got to write out the channel_update).
It's a corner case, but one reported by @darosior and reproduced
on my test node (both with bad gossip_store due to previous iterations
of this patchset!).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This clarifies things a fair bit: we simply add and remove from the
gossip_store directly.
Before this series: (--disable-developer, -Og)
store_load_msec:20669-20902(20822.2+/-82)
vsz_kb:439704-439712(439706+/-3.2)
listnodes_sec:0.890000-1.000000(0.92+/-0.04)
listchannels_sec:11.960000-13.380000(12.576+/-0.49)
routing_sec:3.070000-5.970000(4.814+/-1.2)
peer_write_all_sec:28.490000-30.580000(29.532+/-0.78)
After: (--disable-developer, -Og)
store_load_msec:19722-20124(19921.6+/-1.4e+02)
vsz_kb:288320
listnodes_sec:0.860000-0.980000(0.912+/-0.056)
listchannels_sec:10.790000-12.260000(11.65+/-0.5)
routing_sec:2.540000-4.950000(4.262+/-0.88)
peer_write_all_sec:17.570000-19.500000(18.048+/-0.73)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We have a problem: if we get halfway through writing the compacted store
and run out of disk space, we've already changed half the indexes.
This changes it so we do nothing until writing is finished: then we
iterate through and update indexes. It also weans us off broadcast
ordering, which we can now eliminated.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We didn't count some records before, so we could compare the two counters.
This is much simpler, and avoids reliance on bs.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
(We don't increment the gossip_store version, since there are only a
few commits since the last time we did this).
This lets the reader simply filter messages; this is especially nice since
the channel_announcement timestamp is *derived*, not in the actual message.
This also creates a 'struct gossip_hdr' which makes the code a bit
clearer.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Keeping the uintmap ordering all the broadcastable messages is expensive:
130MB for the million-channels project. But now we delete obsolete entries
from the store, we can have the per-peer daemons simply read that sequentially
and stream the gossip itself.
This is the most primitive version, where all gossip is streamed;
successive patches will bring back proper handling of timestamp filtering
and initial_routing_sync.
We add a gossip_state field to track what's happening with our gossip
streaming: it's initialized in gossipd, and currently always set, but
once we handle timestamps the per-peer daemon may do it when the first
filter is sent.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
