Pluggable session storage is the storage which implemented by the user. For example, it could be used to implement High-Availability (HA) Solutions. One can implement a Pluggable Session Storage, which stores data to shareable storage. If one server with the FIX Engine fails and another reserve server stands up, the FIX Engine can restore session states from the same shareable storage. Such storage should implement the OnixS::FIX::ISessionStorage interface.
The custom implementation must define its persistence, flush, failure, and restart behavior. OnixS::FIX::Session::flushSessionStorage calls OnixS::FIX::ISessionStorage::flush, but does not pass its syncData argument to the custom implementation. If the storage requires device synchronization, define whether OnixS::FIX::ISessionStorage::flush performs it. Compare these requirements in Selecting Storage for Session Messages Logging and apply the implementation's recovery contract alongside the procedures in Restoring Session State from Log Files.
There are few basic rules that should be applied to the implementation of Pluggable Storage.
The user should manage the lifecycle of the storage
The Engine core treats the instance of Pluggable Storage as a "foreign" object and, therefore, never attempts to delete it. On the other hand, there is a strong requirement to keep the storage instance valid until the OnixS::FIX::Session instance is destroyed.
Be as fast as possible
Since Pluggable Storage is incorporated into the message send/receive route, it directly affects sending/receiving latency and overall throughput of the system. So, to keep high performance, it is needed to handle callbacks as fast as possible.
Prevent against exceptions
Exceptions are thrown from Pluggable Storage callbacks, always immediately break a session. So you usually need to use an exception-free code for callbacks.
Don't make assumptions about parameters lifecycle
Messages and pointers passed to Pluggable Storage callbacks should be completely handled before these callbacks return control to Engine core. Whenever you need to keep such data for separate usage, you should make copies of messages or memory blocks and store them in an appropriate place, but don't keep raw pointers, passed to callbacks.
No specific synchronization is needed
The Engine core manages all synchronization issues by itself, so all Pluggable Storage methods are called sequentially. Therefore, you don't need to implement separate synchronization to ensure that OnixS::FIX::ISessionStorage::storeInbound and OnixS::FIX::ISessionStorage::storeOutbound methods are not called simultaneously. Of course, it is still possible to use synchronizations for other needs.
Track message numbers when the message is logged
Message sequence numbers should be tracked while messages are logged, i.e. during OnixS::FIX::ISessionStorage::storeInbound or OnixS::FIX::ISessionStorage::storeOutbound methods are handled. These numbers should be kept and returned to Engine through OnixS::FIX::ISessionStorage::inSeqNum() and OnixS::FIX::ISessionStorage::outSeqNum() calls for incoming and outgoing messages respectively.
OnixS::FIX::ISessionStorage::inSeqNum(int) and OnixS::FIX::ISessionStorage::outSeqNum(int) are called only during the session startup, initial login, and session shutdown of the processes.
This section explains how different methods of Pluggable Storage are called in different situations.
If an exception is thrown by the user's code on any stage (during ISessionStorage or ISessionListener handling), the session becomes disconnected, and subsequent stages are omitted. Logout message will be sent to the counterparty in dependency on the value of the OnixS::FIX::EngineSettings::sendLogoutOnException setting.
There is a quite schematic illustration for the pluggable storage usage.
Please, check the Pluggable Storage Sample included in the distribution package of the FIX Engine.