Daml-script communication needs to be more robust

[samuel-williams-da]

From Curtis, a description of the problem:

• I was just on a call with GS. They are leery about using Daml Script. One reason is they encountered the following situation:
• Running their Daml / Canton system in the cloud
• Were using Daml Script and, I assume, running it locally or on a server in the cloud
• The connection from the host running Daml Script to the PN broke
• This connection breakage made an unrecoverable situation
• A network (socket) connection breaking is a very normal thing. Does Daml Script have any retry mechanism for this? Are there any other gaps in making that connection robust?
• Another related situation was a JWT token expiring in the middle of a Script execution. What would happen in this case?

Netty retry findings

It appears that daml-script's underlying Netty Channel when using GRPC doesn't use any retry logic. The channel builder LedgerClientChannelConfiguration.builderFor sets up TLS and message sizes, but doesn't setup:

.enableRetry()
.maxRetryAttempts(10)

for example. There is also information here on retry directly in GRPC. I've seen a sentiment online that maxRetryAttempts isn't enough.

[dylant-da]

Will be discussed in grilling, will move then

[dylant-da]

Discussed with Curtis at last grilling, explained that making this robust against failure is very difficult, Curtis said he'd get back to us

Putting on backlog until then