get_block large timeout exceeded values in 3.2

[heifner]

As indicated in #677 /v1/chain/get_block is showing: "deadline 2023-01-31T22:14:07.564 exceeded by 914656us"

Outside of the issues being tracked on #677, get_block is currently not honoring http-max-response-time-ms appropriately. get_block runs for abi-serializer-max-time-ms and then checks http-max-response-time-ms after. This means if abi-serializer-max-time-ms is set to 2000 and http-max-response-time-ms is set to 30, then the get_block call can run for say 1750ms and then fail after the expensive work of the abi serializer has completed. In the above exception it appears that abi serializer ran for around 900ms and then failed after the abi serialization was complete.

First thought might be to use the minimum of http-max-response-time-ms and abi-serializer-max-time-ms for the limit on abi serialization of get_block. However, node operators, I believe would prefer for get_block to honor abi-serializer-max-time-ms and ignore http-max-response-time-ms.

Introduced by #96

[matthewdarwin]

Yes, we (me and Aaron) did indicate that : <<node operators, I believe would prefer for get_block to honor abi-serializer-max-time-ms and ignore http-max-response-time-ms.>>

However, before we start changing things around again I would like to better understand what are all the knobs used for and why they should be set depending the expected use of the node in question. I think this merits a more in-depth discussion than a quick fix. Once we decide how this should look, then we can decide on the appropriate code change.

[heifner]

This was worked and fixed by #677. /v1/chain/get_block now honors the minimum of http-max-response-time-ms and abi-serializer-max-time-ms.

[heifner]

Cat: API