KrahoDB is an open-source database designed to support multi-master replication. It is designed on the top of PostgreSQL, providing bidirectional replication, as well as row filtering.
KrahoDB is a PostgreSQL fork. We provide versions based on PostgreSQL 10, 11 and 12. All code is licensed under PostgreSQL license.
The installation process is the same as PostgreSQL.
Use one of the branches named krahodb-X-Y where X.Y is the PostgreSQL version (branch krahodb-10-10 is based on PostgreSQL 10.10).
The following commands build KrahoDB:
$ wget -c https://github.com/timbira/krahodb/archive/KDB_12_0_1.tar.gz
$ tar zxf KDB_12_0_1.tar.gz
$ cd krahodb-KDB_12_0_1
$ ./configure --prefix=$HOME/krahodb --enable-debug
$ make -s -j 4
$ make -s -j 4 install
$ cd contrib
$ make -s -j 4
$ make -s -j 4 install
Let's suppose we have a setup with 4 nodes: node A (10.20.30.1), node B (10.20.30.2), node C (10.20.30.3), and node D (10.20.30.4). We use the same database (testdb) and the same role (kraho) although they can be different in different nodes (don't forget to adjust permissions).
+------------+ +------------+ +------------+
| B | | A | | C |
| 10.20.30.2 |<-------->| 10.20.30.1 |<-------->| 10.20.30.3 |
+------------+ +------------+ +------------+
^
|
v
+------------+
| D |
| 10.20.30.4 |
+------------+
Setup is the same as PostgreSQL with a little differences that we show above.
Adjust listen_addresses to allow connections from other nodes:
listen_addresses = '*'
All nodes must adjust wal_level in postgresql.conf:
wal_level = logical
Default replication parameters is sufficient for fewer nodes (don't worry with additional parameters if you have less than 10 nodes). However, if you have a dozen of nodes, node A should adjust max_wal_senders, max_replication_slots and max_logical_replication_workers (at least the same as leaf nodes). Additionally, max_worker_processes must be set to accommodate one worker per leaf node.
Node A should receive connections from nodes B, C and D so adjust pg_hba.conf in node A:
host testdb kraho 10.20.30.2/32 md5
host testdb kraho 10.20.30.3/32 md5
host testdb kraho 10.20.30.4/32 md5
Nodes B, C and D should receive connection only from node A so adjust pg_hba.conf:
host testdb kraho 10.20.30.1/32 md5
Let's create database testdb and role kraho in each node.
createuser -s -h 10.20.30.1 -U postgres kraho
createuser -s -h 10.20.30.2 -U postgres kraho
createuser -s -h 10.20.30.3 -U postgres kraho
createuser -s -h 10.20.30.4 -U postgres kraho
createdb -h 10.20.30.1 -U postgres -O kraho testdb
createdb -h 10.20.30.2 -U postgres -O kraho testdb
createdb -h 10.20.30.3 -U postgres -O kraho testdb
createdb -h 10.20.30.4 -U postgres -O kraho testdb
Let's populate some data in node A for testing.
pgbench -i -s 5 -h 10.20.30.1 -U kraho testdb
Let's copy the schema from node A to nodes B, C and D.
pg_dump -s -h 10.20.30.1 -U kraho testdb | psql -f - -h 10.20.30.2 -U kraho testdb
pg_dump -s -h 10.20.30.1 -U kraho testdb | psql -f - -h 10.20.30.3 -U kraho testdb
pg_dump -s -h 10.20.30.1 -U kraho testdb | psql -f - -h 10.20.30.4 -U kraho testdb
Let's create publications to send data from node A to nodes B, C and D. It is important to be separate publications because we will define a different filter to tables. Our row filter will be column bid which means that node B, C and D will only have data for an specific bid.
psql -h 10.20.30.1 -U kraho testdb
testdb=# CREATE PUBLICATION pub_a_to_b FOR TABLE pgbench_branches WHERE (bid = 2), pgbench_tellers WHERE (bid = 2), pgbench_accounts WHERE (bid = 2);
CREATE PUBLICATION
testdb=# CREATE PUBLICATION pub_a_to_c FOR TABLE pgbench_branches WHERE (bid = 3), pgbench_tellers WHERE (bid = 3), pgbench_accounts WHERE (bid = 3);
CREATE PUBLICATION
testdb=# CREATE PUBLICATION pub_a_to_d FOR TABLE pgbench_branches WHERE (bid = 4), pgbench_tellers WHERE (bid = 4), pgbench_accounts WHERE (bid = 4);
CREATE PUBLICATION
Let's create subscriptions in nodes B, C and D. By default, it copies data (from node A).
psql -h 10.20.30.2 -U kraho testdb
testdb=# CREATE SUBSCRIPTION sub_b_from_a CONNECTION 'host=10.20.30.1 user=kraho dbname=testdb' PUBLICATION pub_a_to_b WITH (replication_origin_id = 2, filter_origins = 102);
CREATE SUBSCRIPTION
testdb=# \q
psql -h 10.20.30.3 -U kraho testdb
testdb=# CREATE SUBSCRIPTION sub_c_from_a CONNECTION 'host=10.20.30.1 user=kraho dbname=testdb' PUBLICATION pub_a_to_c WITH (replication_origin_id = 3, filter_origins = 103);
CREATE SUBSCRIPTION
testdb=# \q
psql -h 10.20.30.4 -U kraho testdb
testdb=# CREATE SUBSCRIPTION sub_d_from_a CONNECTION 'host=10.20.30.1 user=kraho dbname=testdb' PUBLICATION pub_a_to_d WITH (replication_origin_id = 4, filter_origins = 104);
CREATE SUBSCRIPTION
testdb=# \q
At this point, replication is configured in one direction: A -> B, A -> C and A -> D. Let's configure the other direction.
Let's create publication to each node (B, C, D). Since those nodes only contain data from one branch, do not filter rows.
psql -h 10.20.30.2 -U kraho test
testdb=# CREATE PUBLICATION pub_b_to_a FOR TABLE pgbench_branches, pgbench_tellers, pgbench_accounts;
CREATE PUBLICATION
testdb=# \q
psql -h 10.20.30.3 -U kraho test
testdb=# CREATE PUBLICATION pub_c_to_a FOR TABLE pgbench_branches, pgbench_tellers, pgbench_accounts;
CREATE PUBLICATION
testdb=# \q
psql -h 10.20.30.4 -U kraho test
testdb=# CREATE PUBLICATION pub_d_to_a FOR TABLE pgbench_branches, pgbench_tellers, pgbench_accounts;
CREATE PUBLICATION
testdb=# \q
Let's create subscriptions in node A. Have in mind that data was already copied from node A to the other nodes. In this case, subscription must indicate that data will not be copied.
psql -h 10.20.30.1 -U kraho testdb
testdb=# CREATE SUBSCRIPTION sub_a_from_b CONNECTION 'host=10.20.30.2 user=kraho dbname=testdb' PUBLICATION pub_b_to_a WITH (replication_origin_id = 102, filter_origins = 2, copy_data = false);
CREATE SUBSCRIPTION
testdb=# CREATE SUBSCRIPTION sub_a_from_c CONNECTION 'host=10.20.30.3 user=kraho dbname=testdb' PUBLICATION pub_c_to_a WITH (replication_origin_id = 103, filter_origins = 3, copy_data = false);
CREATE SUBSCRIPTION
testdb=# CREATE SUBSCRIPTION sub_a_from_d CONNECTION 'host=10.20.30.4 user=kraho dbname=testdb' PUBLICATION pub_d_to_a WITH (replication_origin_id = 104, filter_origins = 4, copy_data = false);
CREATE SUBSCRIPTION
testdb=# \q
Setup is ready. Let's insert some rows in node A.
psql -h 10.20.30.1 -U kraho testdb
testdb=# -- this data will not be replicate to nodes because it does not match row filter
testdb=# INSERT INTO pgbench_branches (bid, bbalance, filler) VALUES(100, 123345, 'KrahoDB');
INSERT 0 1
testdb=# -- this row will be replicated to node B
testdb=# INSERT INTO pgbench_tellers (tid, bid, tbalance, filler) VALUES(23456, 2, 22222, 'KrahoDB - node B');
INSERT 0 1
testdb=# -- this row will be replicated to node C
testdb=# INSERT INTO pgbench_tellers (tid, bid, tbalance, filler) VALUES(34567, 3, 33333, 'KrahoDB - node C');
INSERT 0 1
testdb=# -- this row will be replicated to node D
testdb=# INSERT INTO pgbench_tellers (tid, bid, tbalance, filler) VALUES(45678, 4, 44444, 'KrahoDB - node D');
INSERT 0 1
Let's check in the other nodes if data was replicated.
psql -h 10.20.30.2 -U kraho testdb
testdb=# SELECT * FROM pgbench_tellers WHERE tid = 23456;
tid | bid | tbalance | filler
-------+-----+----------+--------------------------------------------------------------------------------------
23456 | 2 | 22222 | KrahoDB - node B
(1 row)
testdb=# \q
psql -h 10.20.30.3 -U kraho testdb
testdb=# SELECT * FROM pgbench_tellers WHERE tid = 34567;
tid | bid | tbalance | filler
-------+-----+----------+--------------------------------------------------------------------------------------
34567 | 3 | 33333 | KrahoDB - node C
(1 row)
testdb=# \q
psql -h 10.20.30.4 -U kraho testdb
testdb=# SELECT * FROM pgbench_tellers WHERE tid = 45678;
tid | bid | tbalance | filler
-------+-----+----------+--------------------------------------------------------------------------------------
45678 | 4 | 44444 | KrahoDB - node D
(1 row)
testdb=# \q
Let's update rows in nodes B, C and D.
psql -h 10.20.30.2 -U kraho testdb
testdb=# -- this row will be replicated to node A
testdb=# UPDATE pgbench_tellers SET tbalance = -1 * tbalance WHERE tid = 23456;
UPDATE 1
testdb=# \q
psql -h 10.20.30.3 -U kraho testdb
testdb=# -- this row will be replicated to node A
testdb=# UPDATE pgbench_tellers SET tbalance = -1 * tbalance WHERE tid = 34567;
UPDATE 1
testdb=# \q
psql -h 10.20.30.4 -U kraho testdb
testdb=# -- this row will be replicated to node A
testdb=# UPDATE pgbench_tellers SET tbalance = -1 * tbalance WHERE tid = 45678;
UPDATE 1
Let's check in node A if data was replicated.
psql -h 10.20.30.1 -U kraho testdb
testdb=# SELECT * FROM pgbench_tellers WHERE tid IN (23456, 34567, 45678);
testdb=# SELECT * FROM pgbench_tellers WHERE tid IN (23456, 34567, 45678);
tid | bid | tbalance | filler
-------+-----+----------+--------------------------------------------------------------------------------------
23456 | 2 | -22222 | KrahoDB - node B
34567 | 3 | -33333 | KrahoDB - node C
45678 | 4 | -44444 | KrahoDB - node D
(3 rows)
Let's remove inserted row in nodes B, C and D to test the other direction.
psql -h 10.20.30.2 -U kraho testdb
testdb=# DELETE FROM pgbench_tellers WHERE tid = 23456;
DELETE 1
testdb=# \q
psql -h 10.20.30.3 -U kraho testdb
testdb=# DELETE FROM pgbench_tellers WHERE tid = 34567;
DELETE 1
testdb=# \q
psql -h 10.20.30.4 -U kraho testdb
testdb=# DELETE FROM pgbench_tellers WHERE tid = 45678;
DELETE 1
testdb=# \q
Let's check in node A if rows were removed.
psql -h 10.20.30.1 -U kraho testdb
testdb=# SELECT * FROM pgbench_tellers WHERE tid IN (23456, 34567, 45678);
tid | bid | tbalance | filler
-----+-----+----------+--------
(0 rows)
Contributions are welcome!
How can you help us?
- open an issue
- bug report
- feature request
- suggestion
- portability problem
- submit a pull request
Before submitting a particular improvement, open an issue to start a discussion about the feature (specially if it is not a trivial change).
Read the PostgreSQL Coding Conventions. It is what we use for KrahoDB. Also, follow the style of the adjacent code! Remove any spurious whitespace (git diff --color). We tend to use underscores or Camelcase (prefer the former). Add useful comments (explain a behavior). Avoid "opens a file" comments.
KrahoDB uses PostgreSQL license. By posting a patch (issue or pull request), you agree that your patch can be distributed under the PostgreSQL license.