This PR restores the previous Pebble configuration, disabling seek compaction.
This feature is still needed by hash mode archive node, mitigating the
overhead of frequent compaction.
This PR changes the Pebble configurations as below:
- increase the MemTableStopWritesThreshold for handling temporary spike
- decrease the L0CompactionConcurrency and CompactionDebtConcurrency to
scale up compaction readily
This introduces two main changes to Pebble's configuration:
(a) Remove the Bloom filter at Level 6
The Bloom filter is never used at the bottom-most level, so keeping it
serves no purpose. Removing it saves storage without affecting read
performance.
(b) Re-enable read-sampling compaction
Read-sampling compaction was previously disabled in the hash-based
scheme because all data was identified by hashes and basically no data
overwrite. Read sampling compaction makes no sense.
After switching to the path-based scheme, data overwrites are much more
common, making read-sampling compaction beneficial and reasonable to re-enable.
Ensure Database.namespace is initialized in pebble.New(...). Without
this, the write-stall metrics registered in onWriteStallBegin/End are
emitted without the intended namespace prefix, while other Pebble
metrics use the provided constructor parameter. This aligns stall
metrics with the rest of the Pebble metric set and fixes inconsistent
metric naming.
---------
Co-authored-by: rjl493456442 <garyrong0905@gmail.com>
This pull request reduces the threshold for triggering compaction at
level0, leading to less compaction debt. This change is helpful in the
case of heavy write-load, mitigating the case of heavy write stalls
caused by compaction.
closes https://github.com/ethereum/go-ethereum/issues/31830
This pull request adjusts the number of allowed memory tables in Pebble.
Pebble allows configuring an arbitrary number of memory tables to hold
unflushed data. When the current memtable becomes full, it is scheduled
for flushing, and a new memtable is allocated to accept subsequent
writes. However, if too many memtables accumulate and are waiting to be
flushed, subsequent writes will stall.
Originally, only two memtables were configured, each with a size of 512
MB for Ethereum mainnet. While this setup works well under normal
conditions, it becomes problematic under heavy write loads. In such scenarios,
flushing is only triggered when more than 512 MB of data is pending, which may
not be responsive enough. Even worse, if compactions are running
concurrently, flushing memtables can become slow due to the heavy IO
overhead, leading to write stalls across the system.
This pull request tries to mitigate the performance degradation by having
more memory tables but with a smaller size. In this case, the pending
writes can be flushed more smoothly and responsively.
This pull request introduces a SyncKeyValue function to the
ethdb.KeyValueStore
interface, providing the ability to forcibly flush all previous writes
to disk.
This functionality is critical for go-ethereum, which internally uses
two independent
database engines: a key-value store (such as Pebble, LevelDB, or
memoryDB for
testing) and a flat-file–based freezer. To ensure write-order
consistency between
these engines, the key-value store must be explicitly synced before
writing to the
freezer and vice versa.
Fixes
- https://github.com/ethereum/go-ethereum/issues/31405
- https://github.com/ethereum/go-ethereum/issues/29819
The metric always has a value, no need to check for the nil.
Seems this code was first introduced here
054412e335/metrics/meter.go (L45-L48)
As the `nilMeter` was removed, so this check seems is useless.
Signed-off-by: jsvisa <delweng@gmail.com>
This PR modifies how the metrics library handles `Enabled`: previously,
the package `init` decided whether to serve real metrics or just
dummy-types.
This has several drawbacks:
- During pkg init, we need to determine whether metrics are enabled or
not. So we first hacked in a check if certain geth-specific
commandline-flags were enabled. Then we added a similar check for
geth-env-vars. Then we almost added a very elaborate check for
toml-config-file, plus toml parsing.
- Using "real" types and dummy types interchangeably means that
everything is hidden behind interfaces. This has a performance penalty,
and also it just adds a lot of code.
This PR removes the interface stuff, uses concrete types, and allows for
the setting of Enabled to happen later. It is still assumed that
`metrics.Enable()` is invoked early on.
The somewhat 'heavy' operations, such as ticking meters and exp-decay,
now checks the enable-flag to prevent resource leak.
The change may be large, but it's mostly pretty trivial, and from the
last time I gutted the metrics, I ensured that we have fairly good test
coverage.
---------
Co-authored-by: Felix Lange <fjl@twurst.com>
This PR adds `DeleteRange` to `ethdb.KeyValueWriter`. While range
deletion using an iterator can be really slow, `DeleteRange` is natively
supported by pebble and apparently runs in O(1) time (typically 20-30ms
in my tests for removing hundreds of millions of keys and gigabytes of
data). For leveldb and memorydb an iterator based fallback is
implemented. Note that since the iterator method can be slow and a
database function should not unexpectedly block for a very long time,
the number of deleted keys is limited at 10000 which should ensure that
it does not block for more than a second. ErrTooManyKeys is returned if
the range has only been partially deleted. In this case the caller can
repeat the call until it finally succeeds.
* cmd/geth, ethdb/pebble: polish method naming and code comment
* implement db stat for pebble
* cmd, core, ethdb, internal, trie: remove db property selector
* cmd, core, ethdb: fix function description
---------
Co-authored-by: prpeh <prpeh@proton.me>
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
cockroachdb/pebble@422dce9 added Errorf to the Logger interface, this change makes it possible to compile geth with that version of pebble by adding the corresponding method to panicLogger.
This is likely the culprit behind several data corruption issues, e.g. where data has been
written to the freezer, but the deletion from pebble does not go through due to process
crash.
One difference between pebble and leveldb is that the latter returns error when performing Get on a closed database, the former does a panic. This may be triggered during shutdown (see #27237)
This PR changes the pebble driver so we check that the db is not closed already, for several operations. It also adds tests to the db test-suite, so the previously implicit assumption of "not panic:ing at ops on closed database" is covered by tests.
* cmd/utils, node: switch to Pebble as the default db if none exists
* node: fall back to LevelDB on platforms not supporting Pebble
* core/rawdb, node: default to Pebble at the node level
* cmd/geth: fix some tests explicitly using leveldb
* ethdb/pebble: allow double closes, makes tests simpler
