Tries to return the approxmost posible string of a Millisatoshi amount using
various unit representations. The function will round to an effective
number of digits. Default: 3.
```
>>> Millisatoshi("100000sat").to_approx_str()
'0.001btc'
>>> Millisatoshi("100msat").to_approx_str()
'0.1sat'
>>> Millisatoshi("10000000sat").to_approx_str()
'0.1btc'
```
Each destructor2 costs 40 bytes, and struct chan is only 120 bytes. So
this drops our memory usage quite a bit:
MCP bench results change:
-vsz_kb:580004-580016(580006+/-4.8)
+vsz_kb:533148
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This has two effects: most importantly, it avoids the problem where
lightningd creates a 800MB JSON blob in response to listchannels,
which causes OOM on the Raspberry Pi (our previous max allocation was
832MB). This is because lightning-cli can start draining the JSON
while we're filling the buffer, so we end up with a max allocation of
68MB.
But despite being less efficient (multiple queries to gossipd), it
actually speeds things up due to the parallelism:
MCP with -O3 -flto before vs after:
-listchannels_sec:8.980000-9.330000(9.206+/-0.14)
+listchannels_sec:7.500000-7.830000(7.656+/-0.11)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We use raw malloc here, again, to handle the failure cases more easily.
I tested it with this hack, then ran the result through `jq --stream '.'`
before and after to make sure it was the same.
diff --git a/cli/lightning-cli.c b/cli/lightning-cli.c
index f840c0786..d83555a51 100644
--- a/cli/lightning-cli.c
+++ b/cli/lightning-cli.c
@@ -295,6 +295,14 @@ static void oom_dump(int fd, char *resp, size_t resp_len, size_t off)
exit(0);
}
+static void *xrealloc(void *p, size_t len)
+{
+ if (len > 1000000)
+ return NULL;
+ return realloc(p, len);
+}
+#define realloc xrealloc
+
int main(int argc, char *argv[])
{
setup_locale();
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This happened with the 800M JSON for the MCP listchannels on the raspberry
pi, and tal calls abort() by default.
We switch to raw malloc here; we could override the error hook for
tal, but this is neater since we're doing low-level things anyway,
I tested it manually with this patch:
diff --git a/lightningd/json_stream.c b/lightningd/json_stream.c
index cec9f5771..206ba37c0 100644
--- a/lightningd/json_stream.c
+++ b/lightningd/json_stream.c
@@ -43,6 +43,14 @@ static void free_json_stream_membuf(struct json_stream *js)
free(membuf_cleanup(&js->outbuf));
}
+static void *membuf_realloc_hack(struct membuf *mb, void *rawelems,
+ size_t newsize)
+{
+ if (newsize > 1000000000)
+ return NULL;
+ return realloc(rawelems, newsize);
+}
+
struct json_stream *new_json_stream(const tal_t *ctx,
struct command *writer,
struct log *log)
@@ -53,7 +61,7 @@ struct json_stream *new_json_stream(const tal_t *ctx,
js->reader = NULL;
/* We don't use tal here, because we handle failure externally (tal
* helpfully aborts with a msg, which is usually right) */
- membuf_init(&js->outbuf, malloc(64), 64, membuf_realloc);
+ membuf_init(&js->outbuf, malloc(64), 64, membuf_realloc_hack);
tal_add_destructor(js, free_json_stream_membuf);
#if DEVELOPER
js->wrapping = tal_arr(js, jsmntype_t, 0);
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We now have a test blockchain for MCP which has the correct channels,
so this is not needed.
Also fix a benchmark script bug where 'mv "$DIR"/log
"$DIR"/log.old.$$' would fail if you log didn't exist from a previous run.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This brings up an interesting quirk though, in that we report "3
attempts", where we really should have done one.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
1. Allow timers to be freed in their callback.
2. Clarify in header that we have to terminate our timer with timer_finished()
eventually.
3. We don't currently have plugins with more than one outstanding timer, but
it certainly would be possible, so fix in_timer to be a counter.
Suggested-by: @ZmnSCPxj
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This was deeply surprising to me; there's a difference between a value not being
specified, and it being specified as "".
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
New fields don't have to be spelled out twice.
The raw version are called _only, so we don't miss a call
accidentally. We can rename them when we finally deprecated old
fields.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Instead of lightningd telling us when it's ready, we ask it.
This also provides an opportunity to have a plugin hook at this point.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
The original idea is to "tal" channel on the "ctx"(In fact, we'd like to set ctx as "ld").
But we already tal channel on "ld" in new_channel(), so "ctx" is unused.
My raspberry pi node hung up on my other node:
lightning_openingd-... chan #1: Got bad message from gossipd: 0db1
This is because we didn't handle that message in one path.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We check that memcmp *isn't* constant time, but that's only true under
-O2 or above: __OPTIMIZE__ doesn't distinguish.
So we need a finer-grained test. Also reduce verbosity by default.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Unfortuntely we get spurious uninitialized variable warnings with
anything but -O3 or no optimization, so set default CWARNFLAGS
appropriately.
MCP bench results without optimization:
store_load_msec:28509-31001(29206.6+/-9.4e+02)
vsz_kb:580004-580016(580006+/-4.8)
store_rewrite_sec:11.640000-12.730000(11.908+/-0.41)
listnodes_sec:1.790000-1.880000(1.83+/-0.032)
listchannels_sec:21.180000-21.950000(21.476+/-0.27)
routing_sec:2.210000-11.160000(7.126+/-3.1)
peer_write_all_sec:36.270000-41.200000(38.168+/-1.9)
MCP bench with -Og: 22% speedup vs no optimization
store_load_msec:21963-23645(22841+/-6.6e+02)
vsz_kb:579916
store_rewrite_sec:10.080000-10.960000(10.456+/-0.3)
listnodes_sec:1.280000-1.390000(1.338+/-0.047)
listchannels_sec:14.770000-16.080000(15.518+/-0.46)
routing_sec:0.990000-6.660000(3.958+/-2.2)
peer_write_all_sec:29.950000-32.950000(31.138+/-1)
MCP bench with -O2: 31% speedup vs no optimization
store_load_msec:20713-22088(21505.6+/-4.8e+02)
vsz_kb:579928
store_rewrite_sec:9.570000-11.200000(10.192+/-0.54)
listnodes_sec:0.960000-1.090000(1.028+/-0.045)
listchannels_sec:10.400000-11.770000(11.012+/-0.48)
routing_sec:0.300000-3.140000(1.978+/-1.1)
peer_write_all_sec:28.980000-30.310000(29.572+/-0.44)
MCP bench with -O3 -flto: 36% speedup vs no optimization
store_load_msec:19616-20191(19862.6+/-1.9e+02)
vsz_kb:578452
store_rewrite_sec:8.980000-9.960000(9.55+/-0.32)
listnodes_sec:0.920000-1.910000(1.18+/-0.38)
listchannels_sec:8.960000-9.450000(9.206+/-0.16)
routing_sec:0.730000-1.850000(1.438+/-0.42)
peer_write_all_sec:28.090000-29.410000(28.772+/-0.42)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We try to look up the funding tx, but it's already spent that to fund
the channel, so we need txindex if this test is to work reliably.
It's not clear to me why this *ever* worked, but if fails on my new
ThreadRipper build machine with valgrind:
> wallettx = l1.bitcoin.rpc.getrawtransaction(wallettxid, True)
...
E bitcoin.rpc.InvalidAddressOrKeyError: {'code': -5, 'message': 'No such mempool transaction. Use -txindex to enable blockchain transaction queries. Use gettransaction for wallet transactions.'}
/usr/lib/python3/dist-packages/bitcoin/rpc.py:231: InvalidAddressOrKeyError
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Instead of reading the store ourselves, we can just send them an
offset. This saves gossipd a lot of work, putting it where it belongs
(in the daemon responsible for the specific peer).
MCP bench results:
store_load_msec:28509-31001(29206.6+/-9.4e+02)
vsz_kb:580004-580016(580006+/-4.8)
store_rewrite_sec:11.640000-12.730000(11.908+/-0.41)
listnodes_sec:1.790000-1.880000(1.83+/-0.032)
listchannels_sec:21.180000-21.950000(21.476+/-0.27)
routing_sec:2.210000-11.160000(7.126+/-3.1)
peer_write_all_sec:36.270000-41.200000(38.168+/-1.9)
Signficant savings in streaming gossip:
-peer_write_all_sec:48.160000-51.480000(49.608+/-1.1)
+peer_write_all_sec:35.780000-37.980000(36.43+/-0.81)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
These are always handed to subdaemons as a set, so group them. This makes
it easier to add an fd (in the next patch).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We need to store the channel capacity for channel_announcement: hand it
in directly rather than having the gossip_store code do a lookup.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Save some overhead, plus gets us ready for giving subdaemons direct
store access. This is the first time we *upgrade* the gossip_store,
rather than just discarding.
The downside is that we need to add an extra message after each
channel_announcement, containing the channel capacity.
After:
store_load_msec:28337-30288(28975+/-7.4e+02)
vsz_kb:582304-582316(582306+/-4.8)
store_rewrite_sec:11.240000-11.800000(11.55+/-0.21)
listnodes_sec:1.800000-1.880000(1.84+/-0.028)
listchannels_sec:22.690000-26.260000(23.878+/-1.3)
routing_sec:2.280000-9.570000(6.842+/-2.8)
peer_write_all_sec:48.160000-51.480000(49.608+/-1.1)
Differences:
-vsz_kb:582320
+vsz_kb:582316
-listnodes_sec:2.100000-2.170000(2.118+/-0.026)
+listnodes_sec:1.800000-1.880000(1.84+/-0.028)
-peer_write_all_sec:51.600000-52.550000(52.188+/-0.34)
+peer_write_all_sec:48.160000-51.480000(49.608+/-1.1)
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Michael Schmoock
Rusty Russell
darosior
trueptolemy
Sjors Provoost