feat: Storm cache supports millisecond resolution #1011
Conversation
Under continuous failure situations having inFlightTTL and graceInterval be seconds can result in a continuous buildup of load. Consider a situation where a specific branch key will never be successfully added to the storm cache. Under continuous load the storm cache will turn this load from concurrent requests to serial requests. However, if a failure is relatively fast, like 25ms, then this serial handling of load at a minimum of 1 request per second will quickly cause the system to be overwhelmed. In addition to supporting milliseconds, when the a storm cache is empty, it will now prioritize fetching. It will triger a fetch every inFlightTTL or graceInterval whichever is shorter. This fetch will still be serial across all cache identifiers. Also, the requirement that the inFlightTTL be greater than or equal to graceInterval has been removed. This does not make sense in the case where the cache is empty.
...tographicMaterialProviders/dafny/AwsCryptographicMaterialProviders/src/CMCs/StormTracker.dfy
Outdated
Show resolved
Hide resolved
|
@seebees and @texastony, I noticed you are updating the smithy model files. |
|
@seebees and @seebees, I noticed you are updating the smithy model files. |
1 similar comment
|
@seebees and @seebees, I noticed you are updating the smithy model files. |
The goal is to not requre more bounds checking but I can do math in the constructor! Saves updating the tests.
`const sleepMilli` does not translate into Java that lets you call `obj.sleepMilli`
|
@seebees and @texastony, I noticed you are updating the smithy model files. |
|
@seebees and @texastony, I noticed you are updating the smithy model files. |
lucasmcdonald3
temporarily deployed
to
Duvet CI
GitHub Actions
Inactive
lucasmcdonald3
temporarily deployed
to
Duvet CI
GitHub Actions
Inactive
This reverts commit 3e6879f.
|
@seebees and @texastony, I noticed you are updating the smithy model files. |
|
@seebees and @texastony, I noticed you are updating the smithy model files. |
# [1.8.0](v1.7.4...v1.8.0) (2024-11-19) ### Bug Fixes * Drop SelectOpt from MutableMap ([bdb6509](bdb6509)) * Externs ([0bc1f96](0bc1f96)) * formatting ([b608ab8](b608ab8)) * **Python-Release:** Run validate tests from release commit ([41c0c94](41c0c94)) * **Python:** CMCs release lock for unhandled runtime exceptions ([#979](#979)) ([1510b77](1510b77)) * **Python:** return error on interrupted sleep ([#1003](#1003)) ([405cf37](405cf37)) * remove input and output traits on DynamoDB operations ([#1012](#1012)) ([8377acf](8377acf)) * return error on interrupted sleep ([#993](#993)) ([f49460a](f49460a)) * rust CI ([42e39cc](42e39cc)) ### Features * **Rust:** Interop test vectors; bump Dafny to 4.9.0 ([#1004](#1004)) ([a505a30](a505a30)) * Storm cache supports millisecond resolution ([#1011](#1011)) ([6f09d5d](6f09d5d))
* chore(release): 1.8.0 [skip ci] # [1.8.0](v1.7.4...v1.8.0) (2024-11-19) ### Bug Fixes * Drop SelectOpt from MutableMap ([bdb6509](bdb6509)) * Externs ([0bc1f96](0bc1f96)) * formatting ([b608ab8](b608ab8)) * **Python-Release:** Run validate tests from release commit ([41c0c94](41c0c94)) * **Python:** CMCs release lock for unhandled runtime exceptions ([#979](#979)) ([1510b77](1510b77)) * **Python:** return error on interrupted sleep ([#1003](#1003)) ([405cf37](405cf37)) * remove input and output traits on DynamoDB operations ([#1012](#1012)) ([8377acf](8377acf)) * return error on interrupted sleep ([#993](#993)) ([f49460a](f49460a)) * rust CI ([42e39cc](42e39cc)) ### Features * **Rust:** Interop test vectors; bump Dafny to 4.9.0 ([#1004](#1004)) ([a505a30](a505a30)) * Storm cache supports millisecond resolution ([#1011](#1011)) ([6f09d5d](6f09d5d))
* chore: Rust implementation for #1011
Under continuous failure situations having inFlightTTL and graceInterval be seconds can result in a continuous buildup of load.
The Storm Cache now supports milliseconds wait times.
When the a storm cache is empty it will also
only wait for a maximum inFlightTTL.
This means that the maximum number of serial requests
that will be waited for
before an error is now
graceInterval / inFlightTTL.In this situation, the cache will return/throw
InFlightTTLExceeded.Consider a situation where a specific branch key
will never be successfully added to the storm cache.
Under continuous load the storm cache will turn
this load from concurrent requests to serial requests.
However, if a failure is relatively fast, like 25ms,
then this serial handling of load
at a minimum of 1 request per second
will quickly cause the system to be overwhelmed.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.