]> git.bitcoin.ninja Git - rapid-gossip-sync-server/log
rapid-gossip-sync-server
6 months agoOnly consider channel announcements with bidirectional updates.
Arik Sosman [Fri, 10 May 2024 23:42:20 +0000 (16:42 -0700)]
Only consider channel announcements with bidirectional updates.

6 months agoIncrease reminder snapshot scope interval threshold.
Arik Sosman [Fri, 10 May 2024 16:41:30 +0000 (09:41 -0700)]
Increase reminder snapshot scope interval threshold.

6 months agoUse reference timestamp for reminder calculation.
Arik Sosman [Fri, 10 May 2024 01:55:54 +0000 (18:55 -0700)]
Use reference timestamp for reminder calculation.

6 months agoAllow passing a timestamp override to snapshot calculation.
Arik Sosman [Fri, 10 May 2024 01:40:37 +0000 (18:40 -0700)]
Allow passing a timestamp override to snapshot calculation.

6 months agoMake reminder inclusion more conservative.
Arik Sosman [Thu, 9 May 2024 23:18:10 +0000 (16:18 -0700)]
Make reminder inclusion more conservative.

6 months agoUpdate test to reflect new reminder mechanics.
Arik Sosman [Thu, 9 May 2024 06:58:52 +0000 (23:58 -0700)]
Update test to reflect new reminder mechanics.

6 months agoDetect beginning of static channel status streaks.
Arik Sosman [Thu, 9 May 2024 06:58:29 +0000 (23:58 -0700)]
Detect beginning of static channel status streaks.

Previously, whenever a channel hasn't received updates in
6+ days, we would automatically send incremental reminders
that only contain the update flags.

However, if a channel has been received updates
within that six-day-timeframe, but all those updates were
identical, it would result in no updates being added to the
snapshot because the mutation set ended up empty and that
data would get purged from the serialization.

In order to avoid the reminder logic being duped by
channels simply being consistent, we now look up the
beginning of the latest continuous stretch of non-mutating
channel updates. If a channel's details have not been
altered in more than six days, we now send reminders no
matter the frequency with which channel updates have been
received since.

6 months agoTest that channels with non-changing updates result in missing reminders.
Arik Sosman [Wed, 8 May 2024 08:37:23 +0000 (01:37 -0700)]
Test that channels with non-changing updates result in missing reminders.

6 months agoDon't panic while panicking.
Arik Sosman [Wed, 24 Apr 2024 23:51:57 +0000 (16:51 -0700)]
Don't panic while panicking.

6 months agoStop tracking Cargo.lock
Arik Sosman [Wed, 8 May 2024 08:03:46 +0000 (01:03 -0700)]
Stop tracking Cargo.lock

8 months agoMerge pull request #74 from TheBlueMatt/main
Arik [Sun, 24 Mar 2024 01:01:04 +0000 (18:01 -0700)]
Merge pull request #74 from TheBlueMatt/main

Bump LDK to 0.0.121/rust-bitcoin 0.30

8 months agoFix unused `Result` in tests
Matt Corallo [Sat, 23 Mar 2024 21:24:38 +0000 (21:24 +0000)]
Fix unused `Result` in tests

8 months agoFix indentation in `test_ln_peers`
Matt Corallo [Sat, 23 Mar 2024 21:24:29 +0000 (21:24 +0000)]
Fix indentation in `test_ln_peers`

8 months agoBump LDK to 0.0.121/rust-bitcoin 0.30, bumping MSRV to 1.63
Matt Corallo [Wed, 13 Mar 2024 14:28:26 +0000 (14:28 +0000)]
Bump LDK to 0.0.121/rust-bitcoin 0.30, bumping MSRV to 1.63

Co-authored-by: Arik Sosman <git@arik.io>
8 months agoMerge pull request #73 from arik-so/arik/2024-03-tokio-runtime-split
Matt Corallo [Fri, 22 Mar 2024 22:44:41 +0000 (22:44 +0000)]
Merge pull request #73 from arik-so/arik/2024-03-tokio-runtime-split

Use separate Tokio runtime for gossip persistence.

8 months agoSimplify client creation.
Arik Sosman [Thu, 21 Mar 2024 00:41:34 +0000 (17:41 -0700)]
Simplify client creation.

8 months agoUse separate tokio runtime for gossip persistence.
Arik Sosman [Tue, 12 Mar 2024 06:41:27 +0000 (23:41 -0700)]
Use separate tokio runtime for gossip persistence.

8 months agoFix CI compilation in beta environment.
Arik Sosman [Thu, 21 Mar 2024 00:45:18 +0000 (17:45 -0700)]
Fix CI compilation in beta environment.

9 months agoMerge pull request #71 from TheBlueMatt/main
Arik [Tue, 6 Feb 2024 19:42:10 +0000 (11:42 -0800)]
Merge pull request #71 from TheBlueMatt/main

Do DB insertions in parallel

10 months agoMerge pull request #72 from lightningdevkit/arik/2024/01/bitcoin-rest-error
Matt Corallo [Thu, 1 Feb 2024 17:39:10 +0000 (17:39 +0000)]
Merge pull request #72 from lightningdevkit/arik/2024/01/bitcoin-rest-error

Improve bitcoind RPC/Rest error message

10 months agoImprove bitcoind RPC/Rest error message.
Arik Sosman [Tue, 30 Jan 2024 19:27:22 +0000 (11:27 -0800)]
Improve bitcoind RPC/Rest error message.

10 months agoDo DB insertions in parallel
Matt Corallo [Mon, 29 Jan 2024 17:24:32 +0000 (17:24 +0000)]
Do DB insertions in parallel

When inserting new gossip into the DB, we block the LDK peer
handling if we get behind. This is mostly okay, but can cause ping
timeouts and reconnections, which isn't ideal. To limit how often
we should see this, here we move to doing the new gossip insertions
in parallel.

10 months agoMerge pull request #70 from domZippilli/2024/01/load-peers-tolerating-trailing-comma
Matt Corallo [Mon, 29 Jan 2024 17:27:14 +0000 (17:27 +0000)]
Merge pull request #70 from domZippilli/2024/01/load-peers-tolerating-trailing-comma

Ignore trailing commas when parsing peers, and log more on error.

10 months agoIgnore trailing commas when parsing peers, and log more on error.
Dominic Zippilli [Thu, 25 Jan 2024 06:02:54 +0000 (22:02 -0800)]
Ignore trailing commas when parsing peers, and log more on error.

10 months agoMerge pull request #68 from johncantrell97/118
Matt Corallo [Wed, 10 Jan 2024 17:38:14 +0000 (17:38 +0000)]
Merge pull request #68 from johncantrell97/118

bump to ldk 118

11 months agobump to ldk 118
John Cantrell [Tue, 5 Dec 2023 21:42:47 +0000 (16:42 -0500)]
bump to ldk 118

12 months agoMerge pull request #65 from arik-so/announcement-history-tests
Matt Corallo [Sat, 11 Nov 2023 17:27:16 +0000 (17:27 +0000)]
Merge pull request #65 from arik-so/announcement-history-tests

Update-recency-based inclusion logic tests

13 months agoTest announcement inclusion logic.
Arik Sosman [Sat, 4 Nov 2023 04:43:33 +0000 (21:43 -0700)]
Test announcement inclusion logic.

13 months agoInclude old updates when necessary.
Arik Sosman [Sat, 4 Nov 2023 06:05:26 +0000 (23:05 -0700)]
Include old updates when necessary.

When a channel has only recently become bidirectional,
but there has not been a new update in the old direction
since the last sync, the latest update in the old direction
must still be included in full because it is the first time
the full channel is being snapshotted.

13 months agoAllow seen override for announcements.
Arik Sosman [Sat, 4 Nov 2023 03:58:21 +0000 (20:58 -0700)]
Allow seen override for announcements.

13 months agoMerge pull request #63 from arik-so/2023/10/ordering_fix
Matt Corallo [Thu, 19 Oct 2023 19:50:38 +0000 (19:50 +0000)]
Merge pull request #63 from arik-so/2023/10/ordering_fix

Fix update recency issue with expanded test coverage

13 months agoOptimize indices for expired value fix lookup.
Arik Sosman [Thu, 19 Oct 2023 18:52:32 +0000 (11:52 -0700)]
Optimize indices for expired value fix lookup.

13 months agoFix expired value being sent as the latest update.
Arik Sosman [Thu, 19 Oct 2023 18:52:06 +0000 (11:52 -0700)]
Fix expired value being sent as the latest update.

13 months agoReproduce test for expired value being sent as the latest update.
Arik Sosman [Thu, 19 Oct 2023 18:49:27 +0000 (11:49 -0700)]
Reproduce test for expired value being sent as the latest update.

13 months agoCreate snapshot generation test.
Arik Sosman [Wed, 18 Oct 2023 00:37:55 +0000 (17:37 -0700)]
Create snapshot generation test.

13 months agoExtract snapshot generation method.
Arik Sosman [Tue, 17 Oct 2023 20:49:48 +0000 (13:49 -0700)]
Extract snapshot generation method.

13 months agoReplicate Mutiny test scenarios.
Arik Sosman [Mon, 16 Oct 2023 15:27:45 +0000 (08:27 -0700)]
Replicate Mutiny test scenarios.

13 months agoAllow setting custom gossip seen timestamps.
Arik Sosman [Sat, 14 Oct 2023 01:39:27 +0000 (18:39 -0700)]
Allow setting custom gossip seen timestamps.

13 months agoMerge pull request #62 from arik-so/2023/10/ordering_fix
Matt Corallo [Sun, 15 Oct 2023 19:54:49 +0000 (19:54 +0000)]
Merge pull request #62 from arik-so/2023/10/ordering_fix

Create Postgres step in CI.

13 months agoAssert that schema is uninitialized.
Arik Sosman [Sat, 14 Oct 2023 01:01:21 +0000 (18:01 -0700)]
Assert that schema is uninitialized.

13 months agoCreate Postgres step in CI.
Arik Sosman [Fri, 13 Oct 2023 19:54:34 +0000 (12:54 -0700)]
Create Postgres step in CI.

13 months agoMerge pull request #60 from arik-so/2023/08/first_unit_test
Matt Corallo [Thu, 12 Oct 2023 17:35:33 +0000 (17:35 +0000)]
Merge pull request #60 from arik-so/2023/08/first_unit_test

First unit test

13 months agoAssert schema sanitization.
Arik Sosman [Tue, 12 Sep 2023 19:25:09 +0000 (12:25 -0700)]
Assert schema sanitization.

13 months agoUse schema as test logger id.
Arik Sosman [Fri, 8 Sep 2023 17:47:11 +0000 (10:47 -0700)]
Use schema as test logger id.

13 months agoCreate first unit test.
Arik Sosman [Tue, 15 Aug 2023 21:39:12 +0000 (14:39 -0700)]
Create first unit test.

13 months agoUpgrade to LDK 117.
Arik Sosman [Tue, 15 Aug 2023 21:38:46 +0000 (14:38 -0700)]
Upgrade to LDK 117.

14 months agoCreate TestLogger.
Arik Sosman [Tue, 15 Aug 2023 21:38:17 +0000 (14:38 -0700)]
Create TestLogger.

14 months agoCreate test-aware db env vars.
Arik Sosman [Tue, 15 Aug 2023 21:37:25 +0000 (14:37 -0700)]
Create test-aware db env vars.

14 months agoMerge pull request #56 from arik-so/2023/08/connect-to-peers-asynchronously
Matt Corallo [Wed, 6 Sep 2023 23:51:06 +0000 (23:51 +0000)]
Merge pull request #56 from arik-so/2023/08/connect-to-peers-asynchronously

Track gossip data prior to peer connection loop.

14 months agoAlways wait 10 seconds prior to attempting reconnection.
Arik Sosman [Wed, 6 Sep 2023 23:26:59 +0000 (16:26 -0700)]
Always wait 10 seconds prior to attempting reconnection.

14 months agoDon't special-case initial connection attempt.
Arik Sosman [Wed, 6 Sep 2023 04:28:15 +0000 (21:28 -0700)]
Don't special-case initial connection attempt.

14 months agoConnect peers independently.
Arik Sosman [Mon, 28 Aug 2023 21:40:35 +0000 (14:40 -0700)]
Connect peers independently.

Previously, we wouldn't start monitoring the gossip data until all
intended peers had either failed or succeeded to connect.

Instead, we're now spawn each peer connection independently, and only
await the first 5 successful peer connections.

14 months agoMerge pull request #57 from arik-so/2023/08/old-incremental-update-fix
Matt Corallo [Wed, 6 Sep 2023 22:16:26 +0000 (22:16 +0000)]
Merge pull request #57 from arik-so/2023/08/old-incremental-update-fix

Send full updates after old last seen updates.

14 months agoSend full updates after old last seen updates.
Arik Sosman [Tue, 29 Aug 2023 01:01:03 +0000 (18:01 -0700)]
Send full updates after old last seen updates.

Previously, whenever we saw that there was a previous update that a
client would have seen, we simply calculated the delta set based on
which properties have changed, and would most likely send an
incremental update set (excepting the case of a new or newly sent
announcement, in which case all sent updates are full).

However, if the last seen update was old, and there's a chance that
a user may have run RGS since, it is possible that due to the
7-day-backdating-mechanism included on the client, the reference
update would no longer be present.

To fix that, anytime we see that a last seen update is more than six
days old, we automatically include a full update.

15 months agoMerge pull request #59 from arik-so/2023/08/custom-intervals
Matt Corallo [Sat, 2 Sep 2023 16:27:08 +0000 (16:27 +0000)]
Merge pull request #59 from arik-so/2023/08/custom-intervals

Distinction between snapshot and symlink intervals

15 months agoDynamically adjust snapshot scopes.
Arik Sosman [Fri, 1 Sep 2023 21:42:40 +0000 (14:42 -0700)]
Dynamically adjust snapshot scopes.

Previously, we had hard-coded factors for the default snapshot
generation interval, which also served as the minimum snapshot
scope. In this commit, we substitute that with a doubling
mechanism that stops once it reaches or exceeds the
21-day-mark, which can be configured using an additional flag.

15 months agoDistinguish between snapshot and symlink intervals.
Arik Sosman [Thu, 31 Aug 2023 22:59:32 +0000 (15:59 -0700)]
Distinguish between snapshot and symlink intervals.

Snapshots may be generated every day or even every couple days,
but symlinks will be generated at the given symlink granularity.

To avoid gcd complications, all snapshot generation intervals
must be multiples of the symlink granularity interval.

15 months agoAllow the snapshot interval to be configurable in the environment variables
Graham Krizek [Thu, 6 Jul 2023 05:31:24 +0000 (00:31 -0500)]
Allow the snapshot interval to be configurable in the environment variables

15 months agoMerge pull request #55 from arik-so/2023/08/multiplication-overflow-fix
Matt Corallo [Mon, 28 Aug 2023 21:20:51 +0000 (21:20 +0000)]
Merge pull request #55 from arik-so/2023/08/multiplication-overflow-fix

Fix multiplication overflow bug

15 months agoFix multiplication overflow bug.
Arik Sosman [Mon, 28 Aug 2023 16:07:19 +0000 (09:07 -0700)]
Fix multiplication overflow bug.

The `snapshot_sync_day_factors` array is sorted
ascendingly, so find() will return on the first
iteration that is at least equal to the requested
interval.

However, the last value in the array is u64::max,
which means that multiplying it with DAY_SECONDS
will overflow. To avoid that, we use saturating_mul.

15 months agoMerge pull request #54 from arik-so/2023/08/incremental-update-fix
Matt Corallo [Thu, 17 Aug 2023 20:55:48 +0000 (20:55 +0000)]
Merge pull request #54 from arik-so/2023/08/incremental-update-fix

Only send full updates with announcements

15 months agoCreate indices for schema version 12.
Arik Sosman [Thu, 17 Aug 2023 20:14:18 +0000 (13:14 -0700)]
Create indices for schema version 12.

15 months agoCreate method for obtaining UTC-prepared Postgres client.
Arik Sosman [Thu, 17 Aug 2023 04:35:55 +0000 (21:35 -0700)]
Create method for obtaining UTC-prepared Postgres client.

15 months agoSend full updates whenever the channel announcement is sent.
Arik Sosman [Thu, 17 Aug 2023 04:50:43 +0000 (21:50 -0700)]
Send full updates whenever the channel announcement is sent.

15 months agoFix ordering of intermediate commits to ensure the latest comes first and is used...
Arik Sosman [Thu, 17 Aug 2023 03:55:51 +0000 (20:55 -0700)]
Fix ordering of intermediate commits to ensure the latest comes first and is used as the authoritative source of what's latest.

15 months agoMerge pull request #53 from arik-so/2023/08/query-timestamp-fix
Matt Corallo [Thu, 17 Aug 2023 00:16:14 +0000 (00:16 +0000)]
Merge pull request #53 from arik-so/2023/08/query-timestamp-fix

Switch timestamp queries to use f64

15 months agoSwitch timestamp reads to use i64.
Arik Sosman [Thu, 17 Aug 2023 00:04:06 +0000 (17:04 -0700)]
Switch timestamp reads to use i64.

15 months agoSwitch timestamp queries to use f64.
Arik Sosman [Wed, 16 Aug 2023 23:18:40 +0000 (16:18 -0700)]
Switch timestamp queries to use f64.

15 months agoMerge pull request #49 from TheBlueMatt/main
Matt Corallo [Sun, 13 Aug 2023 16:37:54 +0000 (16:37 +0000)]
Merge pull request #49 from TheBlueMatt/main

15 months agoMake snapshotter a bit more amenable to changes in interval
Matt Corallo [Sun, 16 Jul 2023 20:41:28 +0000 (20:41 +0000)]
Make snapshotter a bit more amenable to changes in interval

By separating the concept of "seconds in a day" from the snapshot
interval.

15 months agoOnly generate as many symlinks as we need (rather than 27 years worth)
Matt Corallo [Sun, 16 Jul 2023 20:41:00 +0000 (20:41 +0000)]
Only generate as many symlinks as we need (rather than 27 years worth)

15 months agoMerge pull request #51 from arik-so/2023/08/pre_test_refactors
Matt Corallo [Sun, 13 Aug 2023 05:30:42 +0000 (05:30 +0000)]
Merge pull request #51 from arik-so/2023/08/pre_test_refactors

Allow arbitrary logger types.

16 months agoAllow log threshold configuration.
Arik Sosman [Fri, 4 Aug 2023 21:30:02 +0000 (14:30 -0700)]
Allow log threshold configuration.

16 months agoRemove println from verifier.rs
Arik Sosman [Thu, 3 Aug 2023 01:34:01 +0000 (18:34 -0700)]
Remove println from verifier.rs

16 months agoRemove println from tracking.rs
Arik Sosman [Thu, 3 Aug 2023 01:26:55 +0000 (18:26 -0700)]
Remove println from tracking.rs

16 months agoRemove println from snapshot.rs
Arik Sosman [Thu, 3 Aug 2023 01:21:29 +0000 (18:21 -0700)]
Remove println from snapshot.rs

16 months agoRemove println from persistence.rs
Arik Sosman [Thu, 3 Aug 2023 01:19:13 +0000 (18:19 -0700)]
Remove println from persistence.rs

16 months agoRemove println from lookup.rs
Arik Sosman [Thu, 3 Aug 2023 01:16:24 +0000 (18:16 -0700)]
Remove println from lookup.rs

16 months agoRemove println from lib.rs
Arik Sosman [Thu, 3 Aug 2023 01:11:06 +0000 (18:11 -0700)]
Remove println from lib.rs

16 months agoSwitch Logger from Arc to Deref.
Arik Sosman [Wed, 2 Aug 2023 23:00:36 +0000 (16:00 -0700)]
Switch Logger from Arc to Deref.

16 months agoAllow custom logger types.
Arik Sosman [Wed, 2 Aug 2023 22:41:14 +0000 (15:41 -0700)]
Allow custom logger types.

16 months agoUpgrade to 116 release.
Arik Sosman [Wed, 2 Aug 2023 21:33:05 +0000 (14:33 -0700)]
Upgrade to 116 release.

16 months agoMerge pull request #48 from TheBlueMatt/2023-07-further-opt
Arik [Sun, 16 Jul 2023 17:36:13 +0000 (10:36 -0700)]
Merge pull request #48 from TheBlueMatt/2023-07-further-opt

Further optimize intermediate row fetching

16 months agoCorrect upgrade logic to ignore missing indexes 2023-07-further-opt
Matt Corallo [Sun, 16 Jul 2023 17:25:10 +0000 (17:25 +0000)]
Correct upgrade logic to ignore missing indexes

If we upgrade multiple versions in one go, the indexes we're trying
to drop may not exist, and that shouldn't cause us to fail to
upgrade.

16 months agoDrop overly optimistic index
Matt Corallo [Sun, 16 Jul 2023 17:20:56 +0000 (17:20 +0000)]
Drop overly optimistic index

The `channel_updates_id_with_scid_dir_blob` index allows the
intermediate-row-fetching logic to be index-only, but there's very
little reason to do so - we now use subqueries to build the exact
set of rows we want, by id, and then fetch various colums. Having
an index that lets us look up those columns without hitting the
regular table is fine, but there's not a ton of cost to hitting the
table by primary key and maintaining yet another index isn't free.

16 months agoFurther optimize intermediate row fetching by filtering scids first
Matt Corallo [Sun, 16 Jul 2023 17:13:10 +0000 (17:13 +0000)]
Further optimize intermediate row fetching by filtering scids first

Postgresql actually thinks this version is more expensive, but when
I run it its about 4x as fast:

```
ln-gossip=# EXPLAIN ANALYZE SELECT id, direction, blob_signed FROM channel_updates
    WHERE id IN (
        SELECT DISTINCT ON (short_channel_id, direction) id
        FROM channel_updates
        WHERE seen < '2023-07-08 00:00:00' AND short_channel_id IN (
            SELECT DISTINCT ON (short_channel_id) short_channel_id
            FROM channel_updates
            WHERE seen >= '2023-07-08 00:00:00'
        )
        ORDER BY short_channel_id ASC, direction ASC, seen DESC
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Nested Loop  (cost=3771382.74..3772300.66 rows=172186 width=145) (actual time=25270.727..34954.993 rows=129921 loops=1)
   ->  HashAggregate  (cost=3771382.18..3771384.18 rows=200 width=4) (actual time=25267.060..25313.378 rows=129921 loops=1)
         Group Key: channel_updates_1.id
         Batches: 1  Memory Usage: 12321kB
         ->  Unique  (cost=189510.13..3769229.85 rows=172186 width=21) (actual time=693.154..25177.896 rows=129921 loops=1)
               ->  Merge Join  (cost=189510.13..3480148.53 rows=57816264 width=21) (actual time=693.152..23970.052 rows=31781122 loops=1)
                     Merge Cond: (channel_updates_1.short_channel_id = channel_updates_2.short_channel_id)
                     ->  Index Only Scan using channel_updates_scid_dir_seen_desc_with_id on channel_updates channel_updates_1  (cost=0.56..2560497.30 rows=57816264 width=21) (actual time=0.154..19198.277 rows=57605314 loops=1)
                           Index Cond: (seen < '2023-07-08 00:00:00'::timestamp without time zone)
                           Heap Fetches: 0
                     ->  Unique  (cost=189509.57..195871.76 rows=86093 width=8) (actual time=594.634..787.364 rows=68181 loops=1)
                           ->  Sort  (cost=189509.57..192690.67 rows=1272439 width=8) (actual time=594.621..722.739 rows=1452420 loops=1)
                                 Sort Key: channel_updates_2.short_channel_id
                                 Sort Method: external merge  Disk: 17664kB
                                 ->  Index Only Scan using channel_updates_seen_scid on channel_updates channel_updates_2  (cost=0.56..43091.07 rows=1272439 width=8) (actual time=3.461..348.146 rows=1502697 loops=1)
                                       Index Cond: (seen >= '2023-07-08 00:00:00'::timestamp without time zone)
                                       Heap Fetches: 40000
   ->  Index Only Scan using channel_updates_id_with_scid_dir_blob on channel_updates  (cost=0.56..4.60 rows=1 width=145) (actual time=0.074..0.074 rows=1 loops=129921)
         Index Cond: (id = channel_updates_1.id)
         Heap Fetches: 0
 Planning Time: 20.757 ms
 JIT:
   Functions: 18
   Options: Inlining true, Optimization true, Expressions true, Deforming true
   Timing: Generation 0.770 ms, Inlining 51.503 ms, Optimization 25.848 ms, Emission 21.086 ms, Total 99.208 ms
 Execution Time: 34995.683 ms
(26 rows)

ln-gossip=# EXPLAIN ANALYZE SELECT id, direction, blob_signed FROM channel_updates
    WHERE id IN (
        SELECT DISTINCT ON (short_channel_id, direction) id
        FROM channel_updates
        WHERE seen < '2023-07-08 00:00:00'
        ORDER BY short_channel_id ASC, direction ASC, seen DESC
    ) AND short_channel_id IN (
        SELECT DISTINCT ON (short_channel_id) short_channel_id
        FROM channel_updates
        WHERE seen >= '2023-07-08 00:00:00'
    );
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Hash Join  (cost=3049540.37..3050910.30 rows=172186 width=145) (actual time=18697.102..123187.066 rows=129921 loops=1)
   Hash Cond: (channel_updates.short_channel_id = channel_updates_2.short_channel_id)
   ->  Nested Loop  (cost=2851731.51..2852649.44 rows=172186 width=153) (actual time=18029.529..122265.952 rows=393938 loops=1)
         ->  HashAggregate  (cost=2851730.95..2851732.95 rows=200 width=4) (actual time=18027.857..18309.986 rows=393938 loops=1)
               Group Key: channel_updates_1.id
               Batches: 5  Memory Usage: 16449kB  Disk Usage: 7104kB
               ->  Result  (cost=0.56..2849578.62 rows=172186 width=21) (actual time=0.070..17851.034 rows=393938 loops=1)
                     ->  Unique  (cost=0.56..2849578.62 rows=172186 width=21) (actual time=0.044..17809.361 rows=393938 loops=1)
                           ->  Index Only Scan using channel_updates_scid_dir_seen_desc_with_id on channel_updates channel_updates_1  (cost=0.56..2560497.30 rows=57816264 width=21) (actual time=0.043..15596.783 rows=57605314 loops=1)
                                 Index Cond: (seen < '2023-07-08 00:00:00'::timestamp without time zone)
                                 Heap Fetches: 0
         ->  Index Only Scan using channel_updates_id_with_scid_dir_blob on channel_updates  (cost=0.56..4.60 rows=1 width=153) (actual time=0.263..0.263 rows=1 loops=393938)
               Index Cond: (id = channel_updates_1.id)
               Heap Fetches: 0
   ->  Hash  (cost=196732.69..196732.69 rows=86093 width=8) (actual time=662.832..662.833 rows=70509 loops=1)
         Buckets: 131072  Batches: 1  Memory Usage: 3779kB
         ->  Unique  (cost=189509.57..195871.76 rows=86093 width=8) (actual time=542.007..657.269 rows=70509 loops=1)
               ->  Sort  (cost=189509.57..192690.67 rows=1272439 width=8) (actual time=542.004..610.886 rows=1502697 loops=1)
                     Sort Key: channel_updates_2.short_channel_id
                     Sort Method: external merge  Disk: 17664kB
                     ->  Index Only Scan using channel_updates_seen_scid on channel_updates channel_updates_2  (cost=0.56..43091.07 rows=1272439 width=8) (actual time=67.791..274.989 rows=1502697 loops=1)
                           Index Cond: (seen >= '2023-07-08 00:00:00'::timestamp without time zone)
                           Heap Fetches: 39999
 Planning Time: 0.324 ms
 JIT:
   Functions: 23
   Options: Inlining true, Optimization true, Expressions true, Deforming true
   Timing: Generation 0.976 ms, Inlining 8.080 ms, Optimization 37.602 ms, Emission 26.389 ms, Total 73.048 ms
 Execution Time: 123206.373 ms
(29 rows)
```

16 months agoEnforce panic = abort in all build modes
Matt Corallo [Sun, 16 Jul 2023 03:24:53 +0000 (03:24 +0000)]
Enforce panic = abort in all build modes

16 months agoUpgrade to LDK 0.0.116-alpha1
Matt Corallo [Sun, 16 Jul 2023 03:23:34 +0000 (03:23 +0000)]
Upgrade to LDK 0.0.116-alpha1

This includes a fix for the deadlock in #32.

16 months agoDon't hold the `NetworkGraph` read lock across an await point
Matt Corallo [Sun, 16 Jul 2023 03:20:52 +0000 (03:20 +0000)]
Don't hold the `NetworkGraph` read lock across an await point

Holding the `NetworkGraph` read lock across a query await point
can cause a deadlock if another task tries to handle a gossip
message at the same time.

16 months agoSwitch to streaming queries
Matt Corallo [Sun, 16 Jul 2023 00:37:05 +0000 (00:37 +0000)]
Switch to streaming queries

In order to use streaming queries we have to use `tokio-postgres`'s
`query_raw` command, rather than `query`. This should reduce our
memory footprint from 10+GB to well under one.

16 months agoBump tokio to latest LTS version now that we have a higher MSRV.
Matt Corallo [Sun, 16 Jul 2023 00:34:55 +0000 (00:34 +0000)]
Bump tokio to latest LTS version now that we have a higher MSRV.

16 months agoSwitch to Rust edition 2021
Matt Corallo [Sun, 16 Jul 2023 00:32:01 +0000 (00:32 +0000)]
Switch to Rust edition 2021

This increases the MSRV to 1.56, which is now almost two years old,
and lets us use array::IntoIter in the next commit.

16 months agoRemove unused `consider_intermediate_updates` flag..optimizing query
Matt Corallo [Sat, 15 Jul 2023 07:01:16 +0000 (07:01 +0000)]
Remove unused `consider_intermediate_updates` flag..optimizing query

The `consider_intermediate_updates` flag is always set, and must be
set for correctness, so we remove it. Further, we optimize the
query that hung on it somewhat by removing an uneccessary
`ORDER BY` clause which was only neccessary if
`consider_intermediate_updates` were unset.

16 months agoSubstantially optimize reference-row-fetching
Matt Corallo [Sat, 15 Jul 2023 06:42:33 +0000 (06:42 +0000)]
Substantially optimize reference-row-fetching

By first fetching the rows we need from a smaller index, we avoid
walking a large index which contained the full `blob_signed`. This
reduces reference-row-fetching from 680 seconds to 152 seconds when
searching today for reference rows against 7 days ago.

Old:

```
ln-gossip=# EXPLAIN ANALYZE SELECT DISTINCT ON (short_channel_id, direction) id, blob_signed, direction
        FROM channel_updates
        WHERE seen < '2023-07-07 00:00:00' AND short_channel_id IN (
            SELECT DISTINCT ON (short_channel_id) short_channel_id
            FROM channel_updates
            WHERE seen >= '2023-07-07 00:00:00'
        )
        ORDER BY short_channel_id ASC, direction ASC, seen DESC;
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Unique  (cost=186279.46..11921204.82 rows=168910 width=161) (actual time=732.365..680504.173 rows=129985 loops=1)
   ->  Merge Join  (cost=186279.46..11632998.93 rows=57641177 width=161) (actual time=732.364..679193.755 rows=31714061 loops=1)
         Merge Cond: (channel_updates.short_channel_id = channel_updates_1.short_channel_id)
         ->  Index Only Scan using channel_updates_scid_dir_seen on channel_updates  (cost=0.56..10718853.69 rows=57641177 width=161) (actual time=0.638..673675.749 rows=57408667 loops=1)
               Index Cond: (seen < '2023-07-07 00:00:00'::timestamp without time zone)
               Heap Fetches: 0
         ->  Unique  (cost=186278.90..192574.84 rows=84455 width=8) (actual time=478.881..750.241 rows=68210 loops=1)
               ->  Sort  (cost=186278.90..189426.87 rows=1259188 width=8) (actual time=478.878..653.035 rows=1452661 loops=1)
                     Sort Key: channel_updates_1.short_channel_id
                     Sort Method: external merge  Disk: 17680kB
                     ->  Index Only Scan using channel_updates_seen_scid on channel_updates channel_updates_1  (cost=0.56..41481.08 rows=1259188 width=8) (actual time=0.885..264.333 rows=1504495 loops=1)
                           Index Cond: (seen >= '2023-07-07 00:00:00'::timestamp without time zone)
                           Heap Fetches: 2273
 Planning Time: 0.164 ms
 JIT:
   Functions: 9
   Options: Inlining true, Optimization true, Expressions true, Deforming true
   Timing: Generation 21.265 ms, Inlining 37.914 ms, Optimization 113.040 ms, Emission 101.901 ms, Total 274.121 ms
 Execution Time: 680601.155 ms
(19 rows)
```

New:

```
ln-gossip=# EXPLAIN ANALYZE SELECT id, direction, blob_signed FROM channel_updates
    WHERE id IN (
        SELECT DISTINCT ON (short_channel_id, direction) id
        FROM channel_updates
        WHERE seen < '2023-07-07 00:00:00'
        ORDER BY short_channel_id ASC, direction ASC, seen DESC
    ) AND short_channel_id IN (
        SELECT DISTINCT ON (short_channel_id) short_channel_id
        FROM channel_updates
        WHERE seen >= '2023-07-07 00:00:00'
    );
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Hash Join  (cost=2942503.92..2943867.77 rows=169870 width=145) (actual time=22862.627..152436.685 rows=130116 loops=1)
   Hash Cond: (channel_updates.short_channel_id = channel_updates_2.short_channel_id)
   ->  Nested Loop  (cost=2738282.26..2739200.18 rows=169870 width=153) (actual time=22141.452..151504.140 rows=393250 loops=1)
         ->  HashAggregate  (cost=2738281.69..2738283.69 rows=200 width=4) (actual time=22139.440..22339.035 rows=393250 loops=1)
               Group Key: channel_updates_1.id
               Batches: 1  Memory Usage: 45089kB
               ->  Result  (cost=0.56..2736158.32 rows=169870 width=21) (actual time=0.102..21984.409 rows=393250 loops=1)
                     ->  Unique  (cost=0.56..2736158.32 rows=169870 width=21) (actual time=0.074..21943.089 rows=393250 loops=1)
                           ->  Index Only Scan using channel_updates_scid_dir_seen_desc_with_id on channel_updates channel_updates_1  (cost=0.56..2448011.03 rows=57629457 width=21) (actual time=0.073..19776.181 rows=57408667 loops=1)
                                 Index Cond: (seen < '2023-07-07 00:00:00'::timestamp without time zone)
                                 Heap Fetches: 0
         ->  Index Only Scan using channel_updates_id_with_scid_dir_blob on channel_updates  (cost=0.56..4.60 rows=1 width=153) (actual time=0.328..0.328 rows=1 loops=393250)
               Index Cond: (id = channel_updates_1.id)
               Heap Fetches: 0
   ->  Hash  (cost=203159.97..203159.97 rows=84935 width=8) (actual time=721.105..721.107 rows=70731 loops=1)
         Buckets: 131072  Batches: 1  Memory Usage: 3787kB
         ->  Unique  (cost=195708.67..202310.62 rows=84935 width=8) (actual time=552.965..713.465 rows=70731 loops=1)
               ->  Sort  (cost=195708.67..199009.65 rows=1320391 width=8) (actual time=552.962..650.323 rows=1537141 loops=1)
                     Sort Key: channel_updates_2.short_channel_id
                     Sort Method: external merge  Disk: 18064kB
                     ->  Index Only Scan using channel_updates_seen_scid on channel_updates channel_updates_2  (cost=0.56..43421.19 rows=1320391 width=8) (actual time=66.736..324.130 rows=1537141 loops=1)
                           Index Cond: (seen >= '2023-07-07 00:00:00'::timestamp without time zone)
                           Heap Fetches: 68
 Planning Time: 0.520 ms
 JIT:
   Functions: 21
   Options: Inlining true, Optimization true, Expressions true, Deforming true
   Timing: Generation 0.643 ms, Inlining 7.055 ms, Optimization 33.167 ms, Emission 25.782 ms, Total 66.648 ms
 Execution Time: 152458.777 ms
(29 rows)
```

16 months agoMerge pull request #43 from TheBlueMatt/main
Arik [Mon, 10 Jul 2023 18:11:00 +0000 (11:11 -0700)]
Merge pull request #43 from TheBlueMatt/main

Require DB insertions to complete in five seconds

16 months agoRequire DB insertions to complete in fifteen seconds
Matt Corallo [Thu, 6 Jul 2023 16:43:05 +0000 (16:43 +0000)]
Require DB insertions to complete in fifteen seconds

For some reason the mainnet server hung, seemingly on the DB
insertion task. This will improve debugging by simply crashing if
an insertion takes longer than five seconds.

17 months agoMerge pull request #42 from TheBlueMatt/main
Arik [Sun, 2 Jul 2023 19:43:01 +0000 (12:43 -0700)]
Merge pull request #42 from TheBlueMatt/main

#39 Followups

17 months agoRedo indexes for new SELECTs and trivially tweak one old query
Matt Corallo [Sun, 2 Jul 2023 18:45:10 +0000 (18:45 +0000)]
Redo indexes for new SELECTs and trivially tweak one old query

17 months agoBuild reminder updates with correct SCID field
Matt Corallo [Sun, 2 Jul 2023 17:17:07 +0000 (17:17 +0000)]
Build reminder updates with correct SCID field

When the reminder updates were added, a dummy `ChannelUpdate` with
a number of zero'd fields were created under the assumption that
the zero'd fields would be ignored downstream when building
serialized updates. However, the SCID field was `assert`'ed on (and
serialized in the update), causing any reminder updates to cause an
assertion panic.

Instead, we do it the Right Way (tm) here and move the
only-sometimes-available fields into the update type enum, ensuring
we can't access "poison" fields downstream.

17 months agoLet postgres filter timestamps rather than doing it in Rust
Matt Corallo [Sun, 2 Jul 2023 16:12:25 +0000 (16:12 +0000)]
Let postgres filter timestamps rather than doing it in Rust