We want `context.TODO()` to be high signal in the code-base. It should
signal clearly that work is required to thread parent context through to
the call-site. So to keep the signal-to-noise ratio high, we remove any
context.TODO() calls from tests since these will never need to be
replace by a parent context.
After this commit, there is only a single context.TODO() left in the
code-base.
In this commit, we make a change that enables the `auto_vacuum =
incremental` pragma for SQLite databases, but only when the database
file is first created. Incremental auto-vacuum allows SQLite to reclaim
unused space within the database file over time, preventing indefinite
growth.
Find and replace all nolint instances refering to the `lll` linter and
replace with `ll` which is the name of our custom version of the `lll`
linter which can be used to ignore log lines during linting.
The next commit will do the configuration of the custom linter and
disable the default one.
In this commit, we remove the global application level lock from the
postgres backend. This lock prevents multiple write transactions from
happening at the same time, and will also block a writer if a read is on
going. Since this lock was added, we know always open DB connections
with the strongest level of concurrency control available:
`LevelSerializable`. In concert with the new auto retry logic, we ensure
that if db transactions conflict (writing the same key/row in this
case), then the tx is retried automatically.
Removing this lock should increase perf for the postgres backend, as now
concurrent write transactions can proceed, being serialized as needed.
Rather then trying to handle concurrency at the application level, we'll
set postgres do its job, with the application only needing to retry as
necessary.
Previously our RPC calls to etcd would hang even in the case of properly
set dial timeouts and even if there was a network partition. To ensure
liveness we need to make sure that calls fail correctly in case of
system failure. To fix this we add a default timeout of 30 seconds to
each etcd RPC call.
In this commit, we update the project and relevant sub-modules to sqldb
v1.0.2. The next step is to tag a new version of kvdb, then update the
main module to use that.
In this commit, we fix a bug that would cause the entire db to shutdown
if hit a panic (since db operations in the main buckets exit with a
panic) while executing a txn call back. This might be a postgres error
we need to check, so we don't want to bail out, and instead want to pass
up the error to the caller so we can retry if needed.
Some sub-systems like btcwallet will return an error from the database,
but they won't properly wrap it. As a result, we were unable to actually
catch the serialization errors in the first place. To work around this,
we'll now attempt to parse the error string directly.
In this commit, we use exhaustive build tags to ensure that we can
always build the `sqlbase` package, independent of the set build tags.
To do this, we move the type declarations _into_ the parsing functions.
This then allows us to create two versions for each db: with the db, and
without it.
To avoid a module tag round trip to get this working, we use a local
replace for now. Once this is merged in, we can do the tag (along side
rc3), then remove the replace.
Bump all build go versions to v1.21.0
Bump the minimum build package version to v1.19.0
Debian "buster" is not longer supported. Security updates have been
discontinued since June 30th 2022. We will build using the latest
version, "bookworm".
Fix the following conflict,
```
go: conflicting replacements for github.com/ulikunitz/xz:
github.com/ulikunitz/xz@v0.5.11
github.com/ulikunitz/xz@v0.5.8
```
* lnwallet: fix log output msg
The log message is off by one.
* htlcswitch: fail channel when revoking it fails.
When the revocation of a channel state fails after receiving a new
CommitmentSigned msg we have to fail the channel otherwise we
continue with an unclean state.
* docs: update release-docs
* htlcswitch: tear down connection if revocation processing fails
If we couldn't revoke due to a DB error, then we want to also tear down
the connection, as we don't want the other party to continue to send
updates. That may lead to de-sync'd state an eventual force close.
Otherwise, the database might be able to recover come the next
reconnection attempt.
* kvdb: use sql.LevelSerializable for all backends
In this commit, we modify the default isolation level to be
`sql.LevelSerializable. This is the strictness isolation type for
postgres. For sqlite, there's only ever a single writer, so this doesn't
apply directly.
* kvdb/sqlbase: add randomized exponential backoff for serialization failures
In this commit, we add randomized exponential backoff for serialization
failures. For postgres, we''ll his this any time a transaction set fails
to be linearized. For sqlite, we'll his this if we have many writers
trying to grab the write lock at time same time, manifesting as a
`SQLITE_BUSY` error code.
As is, we'll retry up to 10 times, waiting a minimum of 50 miliseconds
between each attempt, up to 5 seconds without any delay at all. For
sqlite, this is also bounded by the busy timeout set, which applies on
top of this retry logic (block for busy timeout seconds, then apply this
back off logic).
* docs/release-notes: add entry for sqlite/postgres tx retry
---------
Co-authored-by: ziggie <ziggie1984@protonmail.com>
