cb045dd0e3
* Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * Merge refs/heads/master into v0.10.1 * add in new patch and workspace * update cloners * Handle rewards overflow (#5122) * Refactoring of initial sync (#5096) * implements blocks queue * refactors updateCounter method * fixes deadlock on stop w/o start * refactors updateSchedulerState * more tests on schduler * parseFetchResponse tests * wraps up tests for blocks queue * eod commit * fixes data race in round robin * revamps fetcher * fixes race conditions + livelocks + deadlocks * less verbose output * fixes data race, by isolating critical sections * minor refactoring: resolves blocking calls * implements init-sync queue * udpate fetch/send buffers in blocks fetcher * blockState enum-like type alias * refactors common code into releaseTicket() * better gc * linter * minor fix to round robin * moves original round robin into its own package * adds enableInitSyncQueue flag * fixes issue with init-sync service selection * Update beacon-chain/sync/initial-sync/round_robin.go Co-Authored-By: terence tsao <terence@prysmaticlabs.com> * initsyncv1 -> initsyncold * adds span Co-authored-by: prylabs-bulldozer[bot] <58059840+prylabs-bulldozer[bot]@users.noreply.github.com> Co-authored-by: Raul Jordan <raul@prysmaticlabs.com> Co-authored-by: terence tsao <terence@prysmaticlabs.com> * Handle rewards overflow * Revert "Refactoring of initial sync (#5096)" This reverts commit |
||
---|---|---|
.. | ||
client | ||
server | ||
README.md |
Cluster private key management tool
This is a primative tool for managing and delegating validator private key assigments within the kubernetes cluster.
Design
When a validator pod is initializing within the cluster, it requests a private key for a deposited validator. Since pods are epheremal, scale up/down quickly, there needs to be some service to manage private key allocations, validator deposits, and re-allocations of previously in-use private keys from terminated pods.
Workflow for bootstraping a validator pod
- Request
n
private keys from the pk manager. - If unallocated private keys exist (from previously terminated pods), assign to the requesting pod.
- If there are not at least
n
keys not in use, generate new private keys, and make the deposits on behalf of these newly generated private keys. - Write the key allocations to a persistent datastore and fulfill the request.
- The client uses these private keys to act as deposited validators in the system.
Server
The server manages the private key database, allocates new private keys, makes validator deposits, and fulfills requests from pods for private key allocation.
Database structure
There are two buckets for the server, unallocated keys and allocated keys.
Unallocated keys bucket:
key | value |
---|---|
private key | nil |
Allocated keys bucket:
key | value |
---|---|
pod name | list of private keys |
Key management design
There are two types of operations with regards to private keys:
- Allocate(podName, keys)
- UnallocateAllKeys(podName)
Allocating keys will first check and attempt to recycle existing, unused keys. If there are no unused keys available (or not enough), new keys are deposited.
Unallocating keys happens when a pod is destroyed. This should return all of that's pods' keys to the unallocated keys bucket.
Assignments HTTP Page /assignments
The server exposes an HTTP page which maps pod names to public keys. This may be useful for determining which logs to follow for a given validator.
Client
The client makes the private key request with a given pod name and generates a keystore with the server response.