Edge trigger node updates #3873
Conversation
chelseakomlo
force-pushed
the
r-edge-trigger-node-watcher
branch
from
1313d88
to
f4ba5f6
Compare
chelseakomlo
changed the title
chelseakomlo
force-pushed
the
r-edge-trigger-node-watcher
branch
from
f4ba5f6
to
7b0f18b
Compare
client/client.go
Outdated
|
|
||
| var nodeHasChanged bool | ||
|
|
||
| for name, new_val := range response.Attributes { |
| if first.IOPS != second.IOPS { | ||
| return false | ||
| } | ||
| if len(first.Networks) != len(second.Networks) { |
There was a problem hiding this comment.
This just checks the length?
client/client.go
Outdated
| // watchNodeUpdates blocks until it is edge triggered. Once triggered, | ||
| // it will update the client node copy and re-register the node. | ||
| func (c *Client) watchNodeUpdates() { | ||
| c.logger.Printf("[DEBUG] client: starting process to watch for node updates.") |
client/client_test.go
Outdated
| if changed, _, _ := c.hasNodeChanged(attrHash, metaHash); changed { | ||
| t.Fatalf("Unexpected hash change.") | ||
| } | ||
| // these constants are only defined when nomad_test is enabled, so these fail |
There was a problem hiding this comment.
I fixed that so the comment isn't accurate and you probably don't need the nolint
client/client_test.go
Outdated
| go c1.watchNodeUpdates() | ||
| c1.updateNode() | ||
| // This needs to be directly called as otherwise the client hangs on | ||
| // attempt to register with a server. S[ecifically, retryRegisterNode is |
client/client_test.go
Outdated
| // attempt to register with a server. S[ecifically, retryRegisterNode is | ||
| // blocking | ||
|
|
||
| // test that the client's copy of the node is also updated |
There was a problem hiding this comment.
Capitalize the start of your comments
client/fingerprint_manager_test.go
Outdated
|
|
||
| err := fm.Run() | ||
| require.Nil(err) | ||
| require.NotEqual(0, node.Resources.CPU) |
client/client.go
Outdated
| @@ -1506,6 +1517,9 @@ func (c *Client) watchNodeUpdates(attrHash, metaHash uint64) { | |||
|
|
|||
| c.retryRegisterNode() | |||
There was a problem hiding this comment.
IIUC, If this retryRegister method returns without registering because it timed out, it won't try again until the next node update comes through triggerNodeUpdate. Is that desirable?
There was a problem hiding this comment.
@dadgar and I talked about this. The node should be registered before watching for updates and the functionality that comes below, if I'm understanding this comment correctly.
test update config copy trigger
chelseakomlo
force-pushed
the
r-edge-trigger-node-watcher
branch
from
8950da5
to
21fa3cd
Compare
chelseakomlo
force-pushed
the
r-edge-trigger-node-watcher
branch
from
21fa3cd
to
ac91ee4
Compare
chelseakomlo
force-pushed
the
r-edge-trigger-node-watcher
branch
from
ac91ee4
to
cd5be3b
Compare
client/client.go
Outdated
| @@ -209,6 +212,7 @@ func NewClient(cfg *config.Config, consulCatalog consul.CatalogAPI, consulServic | |||
| allocUpdates: make(chan *structs.Allocation, 64), | |||
| shutdownCh: make(chan struct{}), | |||
| triggerDiscoveryCh: make(chan struct{}), | |||
| triggerNodeUpdate: make(chan struct{}, 64), | |||
client/client.go
Outdated
| return false | ||
| } | ||
| f := second.Networks[i] | ||
| if e != f { |
There was a problem hiding this comment.
This comparison is a pointer comparison and not a value comparison?
client/client.go
Outdated
|
|
||
| hasChanged = false | ||
| syncTicker.Stop() | ||
| syncTicker = time.NewTicker(c.retryIntv(nodeUpdateRetryIntv)) |
client/client_test.go
Outdated
| // our linter without explicit disabling. | ||
| c1 := TestClient(t, func(c *config.Config) { | ||
| c.Options = map[string]string{ | ||
| driver.ShutdownPeriodicAfter: "true", // nolint: varcheck |
remove impossible test case
client/client.go
Outdated
| hasChanged = false | ||
| timer.Reset(c.retryIntv(nodeUpdateRetryIntv)) | ||
| case <-c.triggerNodeUpdate: | ||
| hasChanged = true |
There was a problem hiding this comment.
Why not put all timer logic in this case and then remove the if statement at the top of the timer case and the reset at the bottom:
case <-c.triggerNodeUpdate:
if hasChanged {
continue
}
hasChanged = true
timer.Reset(c.retryIntv(nodeUpdateRetryIntv))
client/client.go
Outdated
| c.logger.Printf("[DEBUG] client: state changed, updating node.") | ||
| case <-timer.C: | ||
| if !hasChanged { | ||
| continue |
There was a problem hiding this comment.
Right now this code could never fire even if there is a node update
|
I'm going to lock this pull request because it has been closed for 120 days ⏳. This helps our maintainers find and focus on the active contributions. |
Refactor to move client update node mechanism to be edge triggered instead of periodic.
See here for original comment: #3781 (diff)