Support is only for non-conditional chains.

Better support for different runtimes added

Conditional chain support added for Explicit chains. This introduces probabilistic chain support to Xanadu.

Added support to save long term metrics to couchDB

Metrics module added for periodic metrics push to log channel

Added different dispatcher side instrumentation. More to be added.

Added support for speculative deployment of function chains. If function MLE path is present then deploy those parts of the path which are  not already running to avoid cascading cold starts.

The probability model generated so far is generative in nature, in the sense that it tries to model the true probability distribution of the incoming requests. After a large number of requests have been received we can expect to have a reasonable MLE model of the probability distribution. We will then use the viterbi algorithm to detect the most likely path to be taken by a request. This can be used to preallocate the resources in the functions chain span to reduce cascading cold starts.
Path detection has been implemented.
TODO: resource pre-allocation

Added an out-of-band request tracker and request probability measurement module

Integrated RM functionalities with new architecture. Tested using dummy_rm

Added a horizontal autoscalar to scale if number of outstanding requests exceed 100 (currently hardcoded) within a 1s time window.

To prevent requests coming in for the same function from starting too many workers, they are grouped together and one worker is started per group.

Pushed loadbalancer to reverseproxy. This transfer of responsibility simplifies the dispatch daemon.
Also changed loadbalancing logic from number of received request to number of open requests

Tested process termination on memory limit exceeded. Need to test it for isolates

Tested process crash on memory limit exceeded. Need to test it for isolates

Added a min-heap based Loadbalancer to balance based on request count.

Added download  / execute capabilities for grunt. Added better support for RM at DM level

Renaming directories to reflect naming conventions

Added Kafka producers to alert DM?DA of executor states. Removed indirect state gathering module from DA

adding RM interfaces with DM

adding infra for long running isolates, pushed port allocation to dispatcher. TODO: reduce datastructure redundancy

Added long running processes and deployment pipeline bypass support

Reverse proxy based container worker working