Basic Usage of python-rocksdb

Open

The most basic open call is

import rocksdb

db = rocksdb.DB("test.db", rocksdb.Options(create_if_missing=True))

A more production ready open can look like this

import rocksdb

opts = rocksdb.Options()
opts.create_if_missing = True
opts.max_open_files = 300000
opts.write_buffer_size = 67108864
opts.max_write_buffer_number = 3
opts.target_file_size_base = 67108864

opts.table_factory = rocksdb.BlockBasedTableFactory(
    filter_policy=rocksdb.BloomFilterPolicy(10),
    block_cache=rocksdb.LRUCache(2 * (1024 ** 3)),
    block_cache_compressed=rocksdb.LRUCache(500 * (1024 ** 2)))

db = rocksdb.DB("test.db", opts)

It assings a cache of 2.5G, uses a bloom filter for faster lookups and keeps more data (64 MB) in memory before writting a .sst file.

About Bytes And Unicode

RocksDB stores all data as uninterpreted byte strings. pyrocksdb behaves the same and uses nearly everywhere byte strings too. In python2 this is the str type. In python3 the bytes type. Since the default string type for string literals differs between python 2 and 3, it is strongly recommended to use an explicit b prefix for all byte string literals in both python2 and python3 code. For example b'this is a byte string'. This avoids ambiguity and ensures that your code keeps working as intended if you switch between python2 and python3.

The only place where you can pass unicode objects are filesytem paths like

To encode this path name, sys.getfilesystemencoding() encoding is used.

Access

Store, Get, Delete is straight forward

# Store
db.put(b"key", b"value")

# Get
db.get(b"key")

# Delete
db.delete(b"key")

It is also possible to gather modifications and apply them in a single operation

batch = rocksdb.WriteBatch()
batch.put(b"key", b"v1")
batch.delete(b"key")
batch.put(b"key", b"v2")
batch.put(b"key", b"v3")

db.write(batch)

Fetch of multiple values at once

db.put(b"key1", b"v1")
db.put(b"key2", b"v2")

ret = db.multi_get([b"key1", b"key2", b"key3"])

# prints b"v1"
print ret[b"key1"]

# prints None
print ret[b"key3"]

Iteration

Iterators behave slightly different than expected. Per default they are not valid. So you have to call one of its seek methods first

db.put(b"key1", b"v1")
db.put(b"key2", b"v2")
db.put(b"key3", b"v3")

it = db.iterkeys()
it.seek_to_first()

# prints [b'key1', b'key2', b'key3']
print list(it)

it.seek_to_last()
# prints [b'key3']
print list(it)

it.seek(b'key2')
# prints [b'key2', b'key3']
print list(it)

There are also methods to iterate over values/items

it = db.itervalues()
it.seek_to_first()

# prints [b'v1', b'v2', b'v3']
print list(it)

it = db.iteritems()
it.seek_to_first()

# prints [(b'key1', b'v1'), (b'key2, b'v2'), (b'key3', b'v3')]
print list(it)

Reversed iteration

it = db.iteritems()
it.seek_to_last()

# prints [(b'key3', b'v3'), (b'key2', b'v2'), (b'key1', b'v1')]
print list(reversed(it))

SeekForPrev (Take the example from https://github.com/facebook/rocksdb/wiki/SeekForPrev):

db.put(b'a1', b'a1_value')
db.put(b'a3', b'a3_value')
db.put(b'b1', b'b1_value')
db.put(b'b2', b'b2_value')
db.put(b'c2', b'c2_value')
db.put(b'c4', b'c4_value')

it = db.iteritems()
it.seek(b'a1')
assertEqual(it.get(), (b'a1', b'a1_value'))
it.seek(b'a3')
assertEqual(it.get(), (b'a3', b'a3_value'))
it.seek_for_prev(b'c4')
assertEqual(it.get(), (b'c4', b'c4_value'))
it.seek_for_prev(b'c3')
assertEqual(it.get(), (b'c2', b'c2_value'))

Snapshots

Snapshots are nice to get a consistent view on the database

self.db.put(b"a", b"1")
self.db.put(b"b", b"2")

snapshot = self.db.snapshot()
self.db.put(b"a", b"2")
self.db.delete(b"b")

it = self.db.iteritems()
it.seek_to_first()

# prints {b'a': b'2'}
print dict(it)

it = self.db.iteritems(snapshot=snapshot)
it.seek_to_first()

# prints {b'a': b'1', b'b': b'2'}
print dict(it)

MergeOperator

Merge operators are useful for efficient read-modify-write operations. For more details see Merge Operator

A python merge operator must either implement the rocksdb.interfaces.AssociativeMergeOperator or rocksdb.interfaces.MergeOperator interface.

The following example python merge operator implements a counter

class AssocCounter(rocksdb.interfaces.AssociativeMergeOperator):
    def merge(self, key, existing_value, value):
        if existing_value:
            s = int(existing_value) + int(value)
            return (True, str(s).encode('ascii'))
        return (True, value)

    def name(self):
        return b'AssocCounter'


opts = rocksdb.Options()
opts.create_if_missing = True
opts.merge_operator = AssocCounter()
db = rocksdb.DB('test.db', opts)

db.merge(b"a", b"1")
db.merge(b"a", b"1")

# prints b'2'
print db.get(b"a")

We provide a set of default operators rocksdb.merge_operators.UintAddOperator and rocksdb.merge_operators.StringAppendOperator:

from rocksdb.merge_operators import UintAddOperator, StringAppendOperator
opts = rocksdb.Options()
opts.create_if_missing = True
# you should also play with StringAppendOperator
opts.merge_operator = UintAddOperator()
db = rocksdb.DB('/tmp/test', opts)
self.db.put(b'a', struct.pack('Q', 5566))
for x in range(1000):
    self.db.merge(b"a", struct.pack('Q', x))
self.assertEqual(5566 + sum(range(1000)), struct.unpack('Q', self.db.get(b'a'))[0])

PrefixExtractor

According to Prefix API a prefix_extractor can reduce IO for scans within a prefix range. A python prefix extractor must implement the rocksdb.interfaces.SliceTransform interface.

The following example presents a prefix extractor of a static size. So always the first 5 bytes are used as the prefix

class StaticPrefix(rocksdb.interfaces.SliceTransform):
    def name(self):
        return b'static'

    def transform(self, src):
        return (0, 5)

    def in_domain(self, src):
        return len(src) >= 5

    def in_range(self, dst):
        return len(dst) == 5

opts = rocksdb.Options()
opts.create_if_missing=True
opts.prefix_extractor = StaticPrefix()

db = rocksdb.DB('test.db', opts)

db.put(b'00001.x', b'x')
db.put(b'00001.y', b'y')
db.put(b'00001.z', b'z')

db.put(b'00002.x', b'x')
db.put(b'00002.y', b'y')
db.put(b'00002.z', b'z')

db.put(b'00003.x', b'x')
db.put(b'00003.y', b'y')
db.put(b'00003.z', b'z')

prefix = b'00002'

it = db.iteritems()
it.seek(prefix)

# prints {b'00002.z': b'z', b'00002.y': b'y', b'00002.x': b'x'}
print dict(itertools.takewhile(lambda item: item[0].startswith(prefix), it))

Backup And Restore

Backup and Restore is done with a separate rocksdb.BackupEngine object.

A backup can only be created on a living database object.

import rocksdb

db = rocksdb.DB("test.db", rocksdb.Options(create_if_missing=True))
db.put(b'a', b'v1')
db.put(b'b', b'v2')
db.put(b'c', b'v3')

Backup is created like this. You can choose any path for the backup destination except the db path itself. If flush_before_backup is True the current memtable is flushed to disk before backup.

backup = rocksdb.BackupEngine("test.db/backups")
backup.create_backup(db, flush_before_backup=True)

Restore is done like this. The two arguments are the db_dir and wal_dir, which are mostly the same.

backup = rocksdb.BackupEngine("test.db/backups")
backup.restore_latest_backup("test.db", "test.db")

Change Memtable Or SST Implementations

As noted here MemtableFactories, RocksDB offers different implementations for the memtable representation. Per default rocksdb.SkipListMemtableFactory is used, but changing it to a different one is veary easy.

Here is an example for HashSkipList-MemtableFactory. Keep in mind: To use the hashed based MemtableFactories you must set rocksdb.Options.prefix_extractor. In this example all keys have a static prefix of len 5.

class StaticPrefix(rocksdb.interfaces.SliceTransform):
    def name(self):
        return b'static'

    def transform(self, src):
        return (0, 5)

    def in_domain(self, src):
        return len(src) >= 5

    def in_range(self, dst):
        return len(dst) == 5


opts = rocksdb.Options()
opts.prefix_extractor = StaticPrefix()
opts.allow_concurrent_memtable_write = False
opts.memtable_factory = rocksdb.HashSkipListMemtableFactory()
opts.create_if_missing = True

db = rocksdb.DB("test.db", opts)
db.put(b'00001.x', b'x')
db.put(b'00001.y', b'y')
db.put(b'00002.x', b'x')

For initial bulk loads the Vector-MemtableFactory makes sense.

opts = rocksdb.Options()
opts.allow_concurrent_memtable_write = False
opts.memtable_factory = rocksdb.VectorMemtableFactory()
opts.create_if_missing = True

db = rocksdb.DB("test.db", opts)

As noted here TableFactories, it is also possible to change the representation of the final data files. Here is an example how to use a ‘PlainTable’.

opts = rocksdb.Options()
opts.table_factory = rocksdb.PlainTableFactory()
opts.create_if_missing = True

db = rocksdb.DB("test.db", opts)

Change Compaction Style

RocksDB has a compaction algorithm called universal. This style typically results in lower write amplification but higher space amplification than Level Style Compaction. See here for more details, https://github.com/facebook/rocksdb/wiki/Rocksdb-Architecture-Guide#multi-threaded-compactions

Here is an example to switch to universal style compaction.

opts = rocksdb.Options()
opts.compaction_style = "universal"
opts.compaction_options_universal = {"min_merge_width": 3}

See here for more options on universal style compaction, rocksdb.Options.compaction_options_universal

Iterate Over WriteBatch

In same cases you need to know, what operations happened on a WriteBatch. The pyrocksdb WriteBatch supports the iterator protocol, see this example.

batch = rocksdb.WriteBatch()
batch.put(b"key1", b"v1")
batch.delete(b'a')
batch.merge(b'xxx', b'value')

for op, key, value in batch:
    print op, key, value

# prints the following three lines
# Put key1 v1
# Delete a
# Merge xxx value