endpoint.com Bucardo: Replication with tiny little goats http://www.flickr.com/photos/kevincollins/ Selena Deckelmann [email protected]
Bucardo
Dec 05, 2014
Bucardo is a replication system for PostgreSQL. It supports both master-slave (to multiple slaves) and master-master replication, and does not require any modifications to PostgreSQL to run. Maintained by Greg Sabino Mullane, its management application (bucardo_ctl) has recently been significantly improved, making initial setup of replication very fast and easy.
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
endpoint.com
Bucardo:Replication with tiny little goats
http://www.flickr.com/photos/kevincollins/
Selena [email protected]
endpoint.com
★ Software Engineer / Postgres Leadhttp://endpoint.com
★ We’re hiring! http://tr.im/EMGj
Hi!
endpoint.com
http://www.telegraph.co.uk/science/science-news/4409958/Extinct-ibex-is-resurrected-by-cloning.html
endpoint.com
What can it do?
http://www.flickr.com/photos/8602783@N06/2294028540
endpoint.com
Master-slave
• Scaling/Redundancy
• UPGRADES (8.x -> 8.4)
• Reporting databases
• Data warehousing
• Now handles SEQUENCES!
endpoint.com
Master-master
• Write between both databases!
• Sync type: SWAP
• Conflict resolution
endpoint.com
Custom code
• Respond to change
• Filter
• Run arbitrary code!
endpoint.com
STORY TIME!
item onhandskis 10
itemskiitemskiitem
skiitemskiitem
skiitemskiitem
skiitemskiitem
skiitemski
on hand = count(item)
endpoint.com
itemski
new!
bucardo
count(items)and other stuff
prod dbreporting db
itemski
on hand+1
new!
custom code
NOTIFY
NO
TIFY
sync
A
sync Bsync C
goat courtesy of: http://www.flickr.com/photos/nuritwilde/394603312
endpoint.com
Strengths
• Drop-in, no changes to Postgres
• Control daemon/DB can run anywhere
• EXTREME async
• No locking of slaves
• Command-line controls rock!
endpoint.com
The Fine Print
• No smooth DDL handling
• Not for failover (aka high availability)
• bloat: pg_listener, q, bucardo_delta, bucardo_track
• No locking of slaves
• No Windows support
• One developer
http://www.flickr.com/photos/ncarey/135901120/
endpoint.comhttp://wiki.postgresql.org/wiki/Replication,_Clustering,_and_Connection_Pooling
endpoint.com
BucardoConcepts
http://www.flickr.com/photos/kendrick/361173694
endpoint.com
Glossary of terms
• replication: copying data from one place to another
• DDL: Data Definition Language (CREATE/DROP/ALTER)
• DML: Data Manipulation Language (SELECT/INSERT/UPDATE/DELETE)
• triggers: bits of code run before or after DML
endpoint.com
Glossary of terms
• goat: a database object (table or sequence)
• herd: table contains unique name for a group of goats
• syncs: replication events (one herd per sync)
endpoint.com
flexible architecture
bucardo control database
master DB slave DB
endpoint.com
flexible architecture
bucardo control database
master DB slave DB
same db cluster
endpoint.com
flexible architecture
bucardo control database
master DB slave DB
same db cluster
endpoint.com
flexible architecture
bucardo control database
slave DBmaster DB
same db cluster
endpoint.com
Installing
http://www.flickr.com/photos/kendrick/361173694
endpoint.com
Bucardo Requirements
• Postgres (8.3 or higher recommended)
• Perl & some packages:
• DBD::Pg 2.0
• DBI 1.51
• DBIx::Safe 1.2.4
• ExtUtils::MakeMaker 6.32
• pl/perlu and pl/pgsql
endpoint.com
Simple set up
BUCARDO CONTROL database setup:bucardo_ctl install
endpoint.com
Simple set up
MASTER setup:bucardo_ctl add db MyDB name=masterbucardo_ctl add all tables herd=all_tablesbucardo_ctl add all sequences herd=all_tables
SLAVE setup:bucardo_ctl add db MyDB name=slave port=6543
endpoint.com
Setting up, cont.
TEST:
bucardo_ctl validate all
REPLICATE:
bucardo_ctl add sync delta type=pushdelta source=all_tables targetdb=slavebucardo_ctl start
(installs triggers!)
endpoint.com
Setting up
• Initiate a one-time-copy: onetimecopy=[1|2]
• Set up ongoing syncs: enabled by default (can set timeouts)
• MANY configuration options
endpoint.com
Admin stuff
• Can set up defaults in .bucardorc
• Only need one ‘bucardo’ database
• Uses the ‘bucardo’ schema in replication sets for ease of admin, removal
endpoint.com
Smooth upgrades!
bucardo_ctl upgrade
• Stop bucardo first (bucardo_ctl stop)
• applies DDL changes, functions
• see also UPGRADE file
endpoint.com
Stuff to watch out for
• search_path for ‘bucardo’ user
• PL/Perlu & pl/pgsql must be installed
• Uses prepared statements, turn off with:
bucardo_ctl update db MyDB server_side_prepares=0
endpoint.com
LIVE DEMO!
• http://github.com/selenamarie/bucardo_examples
endpoint.com
Internals
http://www.flickr.com/photos/chelseagirlphotos
endpoint.com
Internalsbucardo=# \d List of relations Schema | Name | Type | Owner ---------+-------------------------------+----------+--------- bucardo | audit_pid | table | bucardo bucardo | audit_pid_id_seq | sequence | bucardo bucardo | bucardo_config | table | bucardo bucardo | bucardo_custom_trigger | table | bucardo bucardo | bucardo_custom_trigger_id_seq | sequence | bucardo bucardo | bucardo_log_message | table | bucardo bucardo | bucardo_rate | table | bucardo bucardo | customcode | table | bucardo bucardo | customcode_id_seq | sequence | bucardo bucardo | customcode_map | table | bucardo bucardo | db | table | bucardo bucardo | db_connlog | table | bucardo bucardo | dbgroup | table | bucardo bucardo | dbmap | table | bucardo bucardo | goat | table | bucardo bucardo | goat_id_seq | sequence | bucardo bucardo | herd | table | bucardo bucardo | herdmap | table | bucardo bucardo | q | table | bucardo bucardo | sync | table | bucardo bucardo | upgrade_log | table | bucardo(21 rows)
endpoint.com
goatsTable "bucardo.goat" Column | Type |----------------------+--------------------------+ id | integer | db | text | schemaname | text | tablename | text | reltype | text | pkey | text | qpkey | text | pkeytype | text | has_delta | boolean | ping | boolean | ...
endpoint.com
goats
# bucardo_ctl list tablesTable: archive.course_logging DB: odw_master PK: noneTable: archive.messages DB: odw_master PK: noneTable: archive.student_assignments DB: odw_master PK: noneTable: archive.student_courses DB: odw_master PK: noneTable: logging.activity DB: odw_master PK: activity_name (varchar)Table: logging.course_logging DB: odw_master PK: course_logging_id (int4)Table: public.ecash_transaction DB: odw_master PK: trans_id (int4)Table: public.feedback DB: odw_master PK: date (timestamp)Table: public.holidays DB: odw_master PK: customernum|month|day|year (int4|int4|int4|int4)...
endpoint.com
bucardo_ctl list sequencesSequence 63: logging.logging_seqSequence 64: public.trans_id_seqSequence 65: public.num_seqSequence 66: public.upload_seq
endpoint.com
bucardo databasebucardo=# \d List of relations Schema | Name | Type | Owner ---------+-------------------------------+----------+---------... bucardo | db | table | bucardo bucardo | db_connlog | table | bucardo bucardo | dbgroup | table | bucardo bucardo | dbmap | table | bucardo bucardo | goat | table | bucardo bucardo | goat_id_seq | sequence | bucardo bucardo | herd | table | bucardo bucardo | herdmap | table | bucardo bucardo | q | table | bucardo bucardo | sync | table | bucardo bucardo | upgrade_log | table | bucardo(21 rows)
endpoint.com
Status
# bucardo_ctl statusDays back: 3 User: bucardo Database: bucardo Port: 8000 PID of Bucardo MCP: 6876Name Type State PID Last_good Time I/U/D Last_bad Time========+=====+=====+====+=========+=====+=====+========+====a_delta | P |idle |6886|1m4s |0s |0/0/0|unknown | b_delta | P |idle |6885|1m4s |0s |0/0/0|9h32m47s|0s c_delta | P |idle |6887|1m4s |0s |0/0/0|unknown | d_delta | P |idle |6888|54s |0s |0/0/0|2m6s |3s e_delta | P |idle |6890|1m4s |0s |0/0/0|9h32m47s|0s f_delta | P |idle |6889|1m4s |0s |0/0/0|9h32m47s|0s
endpoint.com
# bucardo_ctl status a_deltaDays back: 3 User: bucardo Database: bucardo Port: 8000======================================================================Sync name: a_deltaCurrent state: idle (PID = 6888)Type: pushdeltaSource herd/database: all_a / a_masterTarget database: a_slaveTables in sync: 26Last good: 3m 34s (time to run: 0s)Last good time: Oct 16, 2009 17:52:20 Target: a_slaveIns/Upd/Del: 20 / 0 / 50Last bad: 4m 47s (time to run: 3s)Last bad time: Oct 16, 2009 17:51:08 Target: a_slaveLatest bad reason: ?PID file: /var/run/bucardo/bucardo.ctl.sync.a_delta.pidPID file created: Fri Oct 16 17:52:09 2009Status: activeLimitdbs: 0Priority: 0Checktime: noneOverdue time: 00:00:00Expired time: 00:00:00Stayalive: yes Kidsalive: yesRebuild index: 0 Do_listen: yesPing: yes Makedelta: no Onetimecopy: 0
endpoint.com
How Bucardo replicates
endpoint.com
bucardo_add_delta
bucardo_triggerkick_[sync]
bucardo_delta
INSERT
NOTIFY!
Triggers IN
SERT
, UPD
ATE,
DEL
ETE
endpoint.com
NOTIFY / LISTEN
• Simple interprocess communication
psql# LISTEN scotch;psql# NOTIFY scotch;Asynchronous notification "scotch" received from server process with PID 1337.
• Any Postgres process can listen in
endpoint.com
Actual Triggers! Table "public.test" Column | Type | Modifiers --------+-----------------------+----------- test | character varying(15) | not nullIndexes: "test_pkey" PRIMARY KEY, btree ("test")Triggers: bucardo_add_delta AFTER INSERT OR DELETE OR UPDATE ON test FOR EACH ROW EXECUTE PROCEDURE bucardo_add_delta_role() bucardo_triggerkick_a_delta AFTER INSERT OR DELETE OR UPDATE ON test FOR EACH STATEMENT EXECUTE PROCEDURE bucardo_triggerkick_a_delta()
endpoint.com
TRIGGER
Flow of control
NOTIFY MCP CTLNOTIFY
INSERT INTO “q”
KID
SELECT from “q”UPDATE “q”
REPLICATIONEVENT!
NOTIFY
Bucardo
endpoint.com
How Syncs happen
• NOTIFY vs. timeout vs. kick
• NOTIFY - from master db to MCP
• timeout - controller detects
• kick - manual/from MCP to controller
endpoint.com
Processes!• MCP, CTL, KID
• MCP: master control process
• CTL: controller (track, kick off/kill KIDs)
• KID: sync processes
endpoint.com
Administrativa
• Totally ok to kill KIDs
• CTL cleans up after you
• May need to kill open database handles (for unfinished COPYs, etc)
endpoint.com
Other interesting tables
On your master DB:
• bucardo_delta table has: ID, txn_time
• bucardo_track table: who’s done what?
endpoint.com
Sync types
• Fullcopy. Timeout/kick. TRUNCATE or DELETE, then COPY
• Pushdelta. Trigger. One way master-slave.
• Swap. Trigger. Two-way. Conflict resolution.
endpoint.com
Note about triggers:
• canonical way to disable triggers: ALTER TABLE DISABLE TRIGGER ALL;:(
• pre 8.3 way: UPDATE pg_class..
• SET session_replication_role = ‘replica’;
• Applies to just the current session
• Thanks, Jan!
endpoint.com
Dealing with failure
• Troubleshooting
• Backing up Bucardo itself
• What happens when you lose the Bucardodb?
endpoint.com
Bucardo hooks
• LIVE DATA CHANGES!
• pass in a hashref (ROW or ROWINFO)
• Returned value - bitmapped (a/b or both!)
• Affect the replication event, side effects
• Example: cache invalidation
endpoint.com
Bucardo Hooks
• customcode & customcode_map
• Conflict handlers (SWAP syncs)
• Pure Perl
• Exception - fix and try again
• Pre and post trigger
endpoint.com
Development
• bucardo.org
• mailing lists: http://bucardo.org/wiki/Bucardo_mailing_list
• Release early, often.
• Please submit patches! bug reports! tests!
endpoint.com
Questions?
endpoint.com
Thanks!
http://www.slideshare.net/selenamarie/bucardo
@selenamarie on Twitter
rss: http://www.chesnok.com/daily
Related Documents
