dicetechnology / dice-fairlink Goto Github PK

SPI uses the most appropriate driver (sometimes the first) that accepts a given URL.

The java.sql.Driver interface recommends returning null if a URL is not accepted in the connect method (equivalent to returning falseon the accepts method, depending on the runtime).

dice-fairlink currently behaves according to this specification, but logs an exception if the URL is not accepted (ie, cannot be turned into a ParsedUrl).

This can become quite noisy and is not useful, as it's normal execution rather than an exception as such.

This behaviour should be changed to silently return null on the connect method and false in the accepts method implementations of the java.sql.Driver interface

I am trying to make dice-fairlink library work with HikariCP. First of all, is this something possible? I couldn't find any resources to verify the same.

User may need to force the driver to re-actualise the list of replicas. Need an interface (method?) for that.

Currently little to no documentation exists at the source code level.

At least the public classes and methods should have some javadoc documentation.

Consider the following scenario:

1. Driver queried the cluster for RO replicas' endpoints.
2. N records returned of 'active' state.
3. N records added to the list.
4. X RO replicas went down.
5. Application requests connection to be established.
6. Driver pulls record which points to the replica, which is no longe accessible.
7. Driver tries to open JDBC connection.
8. Driver receives SQLException with IOException as cause.

Need to add logic which will:

1. Handle exception and try to open the connection to the healthy address, respecting the general distribution logic.
2. Optionally: Remove those records from the list or somehow mark it as down.

One use case of a database cluster is by an application with a heavy bias towards reads.
In this case, the master database can be mostly idle

This task is to implement a configurable mechanism where the user specifies whether (/how much) a connection to the writer instance should be returned, additionally to the read round-robin.

Current: jdbc:auroraro:mysql//my-cluster/my-schema
Should be: jdbc:auroraro:mysql://my-cluster/my-schema

For low replicaPollInterval and/or high number of applications using dice-fairlink, it is possible to encounter an undocumented RDS API limit that will result in throttling and eventually in failure to discover cluster members.

This means dice-fairlink will not be able to update the cluster information or, depending on the client application using the library, it may even not be able to start.

This has been encountered with replicaPollInterval set to 30 seconds and with 60 dice-fairlink enabled applications within the same AWS account.

This is both not a dice-fairlink bug nor it is within dice-fairlink possibilities to solve.

However, this severely compromises functionality and an alternative discovery mean must be explored.