btree – simple BTree database

The btree module implements a simple key-value database using external storage (disk files, or in general case, a random-access stream). Keys are stored sorted in the database, and besides efficient retrieval by a key value, a database also supports efficient ordered range scans (retrieval of values with the keys in a given range). On the application interface side, BTree database work as close a possible to a way standard dict type works, one notable difference is that both keys and values must be bytes objects (so, if you want to store objects of other types, you need to serialize them to bytes first).

The module is based on the well-known BerkelyDB library, version 1.xx.


import btree

# First, we need to open a stream which holds a database
# This is usually a file, but can be in-memory database
# using uio.BytesIO, a raw flash partition, etc.
# Oftentimes, you want to create a database file if it doesn't
# exist and open if it exists. Idiom below takes care of this.
# DO NOT open database with "a+b" access mode.
    f = open("mydb", "r+b")
except OSError:
    f = open("mydb", "w+b")

# Now open a database itself
db =

# The keys you add will be sorted internally in the database
db[b"3"] = b"three"
db[b"1"] = b"one"
db[b"2"] = b"two"

# Assume that any changes are cached in memory unless
# explicitly flushed (or database closed). Flush database
# at the end of each "transaction".

# Prints b'two'

# Iterate over sorted keys in the database, starting from b"2"
# until the end of the database, returning only values.
# Mind that arguments passed to values() method are *key* values.
# Prints:
#   b'two'
#   b'three'
for word in db.values(b"2"):

del db[b"2"]

# No longer true, prints False
print(b"2" in db)

# Prints:
#  b"1"
#  b"3"
for key in db:


# Don't forget to close the underlying stream!

Functions, *, flags=0, cachesize=0, pagesize=0, minkeypage=0)

Open a database from a random-access stream (like an open file). All other parameters are optional and keyword-only, and allow to tweak advanced parameters of the database operation (most users will not need them):

  • flags - Currently unused.
  • cachesize - Suggested maximum memory cache size in bytes. For a board with enough memory using larger values may improve performance. The value is only a recommendation, the module may use more memory if values set too low.
  • pagesize - Page size used for the nodes in BTree. Acceptable range is 512-65536. If 0, underlying I/O block size will be used (the best compromise between memory usage and performance).
  • minkeypage - Minimum number of keys to store per page. Default value of 0 equivalent to 2.

Returns a BTree object, which implements a dictionary protocol (set of methods), and some additional methods described below.



Close the database. It’s mandatory to close the database at the end of processing, as some unwritten data may be still in the cache. Note that this does not close underlying stream with which the database was opened, it should be closed separately (which is also mandatory to make sure that data flushed from buffer to the underlying storage).


Flush any data in cache to the underlying stream.

btree.get(key, default=None)
btree.__setitem__(key, val)

Standard dictionary methods.


A BTree object can be iterated over directly (similar to a dictionary) to get access to all keys in order.

btree.keys([start_key[, end_key[, flags]]])
btree.values([start_key[, end_key[, flags]]])
btree.items([start_key[, end_key[, flags]]])

These methods are similar to standard dictionary methods, but also can take optional parameters to iterate over a key sub-range, instead of the entire database. Note that for all 3 methods, start_key and end_key arguments represent key values. For example, values() method will iterate over values corresponding to they key range given. None values for start_key means “from the first key”, no end_key or its value of None means “until the end of database”. By default, range is inclusive of start_key and exclusive of end_key, you can include end_key in iteration by passing flags of btree.INCL. You can iterate in descending key direction by passing flags of btree.DESC. The flags values can be ORed together.



A flag for keys(), values(), items() methods to specify that scanning should be inclusive of the end key.


A flag for keys(), values(), items() methods to specify that scanning should be in descending direction of keys.