Here we adjust the ForEachNodeCached graph DB method to pass in a node's
addresses into the provided call-back if requested. This will allow us
to improve the performance of node/channel iteration in the autopilot
subserver.
In preparation for removing the ForEachNode/ForEachChannel call pattern,
we refactor the autopilot subserver by simplifying the `ChannelEdge`
such that it no longer returns the `Node` type which has a
`ForEachChannel` method on it. The aim is to completely remove the
`ForEachChannel` usage.
Previously, ForEachNodeCached would batch fetch node _feature_ data but
would still fetch the channel set of each node in a node-by-node fashion
which is not ideal. So this commit updates this method to make use of
the new sqldb.ExecuteCollectAndBatchWithSharedDataQuery helper. It lets
us batch load channel data for a range of node IDs.
This _greatly_ improves the performance of the method.
In this commit, we update the SQLStore's ForEachNodeCacheable and
ForEachNodeCached methods to use batch collection for node feature bits.
This results in the following performance gains:
```
name old time/op new time/op delta
ForEachNodeCacheable-native-sqlite-10 184ms ± 2% 145ms ±10% -21.45% (p=0.000 n=10+10)
ForEachNodeCacheable-native-postgres-10 697ms ± 8% 51ms ± 4% -92.68% (p=0.000 n=9+10)
```
In this commit we add a new helper method in the sqldb package:
ExecuteCollectAndBatchWithSharedDataQuery. This can be used to paginate
through items in the database while at the same time performing batch
data collection for those items.
In this commit, we add a makefile helper that can be used to add the
"-short" flag to the "go test" command when running unit tests via "make
unit ...".
Tests can then be expanded with a `testing.Short()` check to potentially
skip the test in short mode. This is useful for if a dev wants to
quickly run most of the tests in a package but would like to opt out of
running the longer form tests such as stress tests or rapid generation
tests.
This commit adds this check to two graph/db tests.
This commit is a pure refactor. Here all we are doing is removing the
old `pageSize` constant used in the SQLStore code and instead using the
value provided in the QueryConfig struct.
We rename this helper along the config types & helper types for it
because the word "page" is used more often in the context of paging
through results using an offset and limit whereas this helper is
specifically used to split up the slice in queries of the form
"WHERE x in []slice". We do this rename so that there is mimimal
confusion in contexts where we use batching along with actual paging.
The config struct is also renamed to QueryConfig in preparation for it
holding more config options.
Here, we add a new `buildNodeWithBatchData` helper method that can be
used to construct a `models.LightningNode` object using pre-fetched
batch data. The existing `buildNode` method is then adjusted to use this
new helper.
In this commit, we add the queries that will be needed to batch-fetch
the data of a set of nodes. The logic for using these new queries is
also added but not used yet.
In this commit, we remove the LEFT JOIN query that was used for fetching
a nodes addresses. The reason it was used before was to ensure that we'd
get an empty address list if the node did exist but had no addresses.
This was for the purposes of the `AddrsForNode` method since it needs to
return false/true to indicate if the given node exists.
This commit adds a helper test that is similar to TestPopulateDBs except
for the fact that it uses the MigrateGraphToSQL function directly to
migrate a local kvdb-sql graph to a native SQL one.
This is a prep commit that just adds all the boilerplate connection
logic for connecting to various graph DBs. These will be used in
upcoming commits which will add tests and benchmarks.
Use the new `SLICES` directive to add a DeleteChannels query which takes
a set of DB channel IDs. Then replace all our calls to DeleteChannel
with a paginated call to DeleteChannels.
This commit adds a new GetChannelsByOutpoints query which takes a slice
of outpoint strings. This lets us then update PruneGraph to use
paginated calls to GetChannelsByOutpoints instead of making one DB call
per outpoint.
Here, a new query (GetChannelsByOutpoints) is added which makes use of
the /*SLICE:outpoints*/ directive & added workaround. This is then used
in a test to demonstrate how the ExecutePagedQuery helper can be used to
wrap a query like this such that calls are done in pages.
The query that has been added will also be used by live code paths in an
upcoming commit.
Along with a test for it. This helper will allow us to easily create a
pagination wrapper for queries that will make use of the new
/*SLICE:<field_name>*/ directive. The next commit will add a test
showing this.
To ensure that the RandNodeAlias helper can be used elsewhere to
generate random aliases, we adjust it in this test to only produce valid
UTF-8 strings as required by the spec.
In preparation for tests where we will want to spin up SQL DBs many
times, we do some refactoring so that it is easy to re-use postgres
fixtures since those are expensive to spin up.
This commit passes no-op reset callbacks to various places where they
are needed in the KVStore. Upcoming commits will replace the no-ops by
adding reset params to the methods that require it.
This commit adds a basic itest that is run under the `test_native_sql`
flag. It tests that the migration is idempotent and that the node cant
be restarted without the `db.use-native-sql` flag once the graph
migration has been run.
This commit plugs in the graph kvdb-to-sql migration for builds
containing the `test_native_sql` tag. This will allow us to perform
local tests and write itests for the migration without exposing it to
the production release build.
This commit expands `MigrateGraphToSQL` to include migration of the
zombie index.
NOTE: we take this opportunity to clean up the zombie index a bit by
first checking for each channel in the zombie index if it has been
marked as closed in the closed-scid index. If it has, then there is no
need to include it in the zombie index. A special case test for this
is added too.
For a kvdb backed invoices DB, we sometimes will return the
ErrNoInviocesCreated error if no invoices have ever been created on the
node. In these cases we should still wrap the returned error in the grpc
NotFound code.
The previous implementation of the graph cache evictor used
time.AfterFunc, which introduced a race condition. The closure
passed to AfterFunc could execute before the call returned and
assigned the timer to the r.graphCacheEvictor field.
This created a scenario where the closure would attempt to call
Reset() on a nil r.graphCacheEvictor, causing a panic.
This commit fixes the race by replacing time.AfterFunc with a
more robust pattern using time.NewTimer and a dedicated goroutine.
The new timer is created and assigned immediately, ensuring that
r.graphCacheEvictor is never nil when accessed.
The dedicated goroutine now safely manages the timer's lifecycle,
resetting it upon firing and stopping it gracefully upon server
shutdown, which also prevents goroutine leaks.
In this commit, the `MigrateGraphToSQL` function is expanded to migrate
the channel and channe policy data. Both of these have the special case
where the kvdb store records may contain invalid TLV. If we encounter a
channel with invalid TLV, we skip it and its policies. If we encounter a
policy with invalid TLV, we skip it.
The `TestMigrateGraphToSQL` and `TestMigrationWithChannelDB` tests are
updated accordingly.
Elle Mouton