Scalable PostgreSQL connection pooler
Go to file
Dmitry Simonenko 47b65e8ef2 odyssey: fix show servers callback arg 2018-12-07 15:28:32 +03:00
cmake odyssey: add support for BoringSSL 2018-11-23 17:07:39 +03:00
debian Debian package infrastructure 2017-08-21 15:18:28 +03:00
documentation odyssey: allow to set process/workers priority 2018-11-14 16:14:13 +03:00
scripts updated version & file structure 2018-05-31 23:52:07 +03:00
sources odyssey: fix show servers callback arg 2018-12-07 15:28:32 +03:00
stress machinarium: rework time update logic 2018-11-27 16:19:16 +03:00
test machinarium: rework msg_cache/coroutine_cache per machine thread 2018-11-19 17:52:37 +03:00
third_party odyssey: major router and shared logic rework 2018-12-06 17:23:15 +03:00
.gitignore odyssey: remove old machinarium tests 2018-04-04 17:40:43 +03:00
.gitmodules odyssey: remove machinarium as submodule 2018-03-29 14:20:50 +03:00
.travis.yml Travis has two cores. Let's use them. 2018-10-02 17:46:52 +03:00
AUTHORS odyssey: update LICENSE and AUTHORS 2018-05-21 17:39:43 +03:00
CMakeLists.txt odyssey: add support for BoringSSL 2018-11-23 17:07:39 +03:00
LICENSE odyssey: update LICENSE and AUTHORS 2018-05-21 17:39:43 +03:00
README.md odyssey: allow to set process/workers priority 2018-11-14 16:14:13 +03:00
odyssey.conf odyssey: allow to set process/workers priority 2018-11-14 16:14:13 +03:00

README.md



Odyssey

Advanced multi-threaded PostgreSQL connection pooler and request router.

Project status

Although we run Odyssey in production, the project is currently in Beta. We appreciate any kind of feedback and contribution to the project.

Design goals and main features

Multi-threaded processing

Odyssey can significantly scale processing performance by specifying a number of additional worker threads. Each worker thread is responsible for authentication and proxying client-to-server and server-to-client requests. All worker threads are sharing global server connection pools. Multi-threaded design plays important role in SSL/TLS performance.

Advanced transactional pooling

Odyssey tracks current transaction state and in case of unexpected client disconnection can emit automatic Cancel connection and do Rollback of abandoned transaction, before putting server connection back to the server pool for reuse. Additionally, last server connection owner client is remembered to reduce a need for setting up client options on each client-to-server assignment.

Better pooling control

Odyssey allows to define connection pools as a pair of Database and User. Each defined pool can have separate authentication, pooling mode and limits settings.

Authentication

Odyssey has full-featured SSL/TLS support and common authentication methods like: md5 and clear text both for client and server authentication. Additionally it allows to block each pool user separately.

Logging

Odyssey generates universally unique identifiers uuid for client and server connections. Any log events and client error responses include the id, which then can be used to uniquely identify client and track actions. Odyssey can save log events into log file and using system logger.

Architecture and internals

Odyssey has sophisticated asynchronous multi-threaded architecture which is driven by custom made coroutine engine: machinarium. Main idea behind coroutine design is to make event-driven asynchronous applications to look and feel like being written in synchronous-procedural manner instead of using traditional callback approach.

One of the main goal was to make code base understandable for new developers and to make an architecture easily extensible for future development.

More information: Architecture and internals.

Build instructions

Currently Odyssey runs only on Linux. Supported platforms are x86/x86_64.

To build you will need:

  • cmake >= 2.8
  • gcc >= 4.6
  • openssl
git clone git://github.com/yandex/odyssey.git
cd odyssey
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make

Configuration reference

Service
Logging
Performance
System
Global limits
Listen
Routing
Storage
Database and user
Architecture and Internals