chore(rfc): operation cache warmer #1115
Conversation
StarpTech
requested review from
devsergiy,
AndreasZeissner,
jensneuse,
thisisnithin,
Aenimus and
JivusAyrus
rfc/distributed-operation-cache.md
Outdated
| The User can push individual operations to the distributed operation cache by using the CLI: | ||
|
|
||
| ```bash | ||
| wgc router cache add -g mygraph operations.json |
There was a problem hiding this comment.
Should this be under the namespace "router"?
At which level do we apply cache warming?
Org? Namespace? Graph? Router?
As an alternative, you could consider creating a "cache-warming-bucket", which you can then use to add operations, and which can be used in the router config, e.g.:
version: "1"
cache_warmup:
enabled: true
interval: 5m
bucket: myBucketI'm not sure though if this is the right direction.
The cache warming bucket is somewhat closely related to a Graph because the containing operations need to be compatible with the client schema of the graph.
rfc/distributed-operation-cache.md
Outdated
| ] | ||
| ``` | ||
|
|
||
| The cli command is idempotent and always updates the cache with the latest operations. This doesn't trigger the computation of the Top-N operations which is done periodically by the Cosmo Platform. |
There was a problem hiding this comment.
Another aspect to think about is pre-compiling operation plans.
We could make plans serializable, generate and serialize them at composition time.
Once the router wants to "load" them, it doesn't have to plan everything but can load the plan much faster.
StarpTech
changed the title
| The User can push individual operations to the operation cache by using the CLI: | ||
|
|
||
| ```bash | ||
| wgc federated-graph operation-cache add --graph mygraph --file operations.json |
There was a problem hiding this comment.
Curious about manual cache invalidation.
What will happen in the following scenario:
First, we run:
wgc federated-graph operation-cache add --graph mygraph --file operationA.json
Then we run:
wgc federated-graph operation-cache add --graph mygraph --file operationB.json
What will we have in the cache after the second operation? Will it be just Operation B or both Operation A and Operation B?
Curious, because if it is both Operation A and Operation B, the cache eventually will be filled with the manual operation and there will not be space for automatic operations unless we can invalidate it explicitly.
There was a problem hiding this comment.
This is right. Both operations will be in the "batch". This is the part where a customer can manage the cache themself and is responsible of it. At some point, we could also introduce a percentage limit about how much space can be reserved for manual pushed operations. In the first version, we want to keep it simple.
|
This PR was marked stale due to lack of activity. It will be closed in 14 days. |
|
Closed as inactive. Feel free to reopen if this PR is still being worked on. |
|
This PR was marked stale due to lack of activity. It will be closed in 14 days. |
|
This PR was marked stale due to lack of activity. It will be closed in 14 days. |
Motivation and Context
TODO