Edge trigger node updates #3873

chelseakomlo · 2018-02-14T22:53:03Z

Refactor to move client update node mechanism to be edge triggered instead of periodic.

See here for original comment: #3781 (diff)

dadgar · 2018-02-22T19:00:21Z

client/client.go

+
+	var nodeHasChanged bool
+
+	for name, new_val := range response.Attributes {


Go style would be newVal

dadgar · 2018-02-22T19:01:27Z

client/client.go

+	if first.IOPS != second.IOPS {
+		return false
+	}
+	if len(first.Networks) != len(second.Networks) {


This just checks the length?

dadgar · 2018-02-22T19:02:59Z

client/client.go

+// 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.")


dadgar · 2018-02-22T19:04:34Z

client/client_test.go

-	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


I fixed that so the comment isn't accurate and you probably don't need the nolint

dadgar · 2018-02-22T19:05:01Z

client/client_test.go

+	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


Spelling error

dadgar · 2018-02-22T19:05:17Z

client/client_test.go

+	// attempt to register with a server. S[ecifically, retryRegisterNode is
+	// blocking
+
+	// test that the client's copy of the node is also updated


Capitalize the start of your comments

dadgar · 2018-02-22T19:06:36Z

client/fingerprint_manager_test.go

+
+	err := fm.Run()
+	require.Nil(err)
+	require.NotEqual(0, node.Resources.CPU)


preetapan · 2018-02-23T00:05:38Z

client/client.go

@@ -1506,6 +1517,9 @@ func (c *Client) watchNodeUpdates(attrHash, metaHash uint64) {

 				c.retryRegisterNode()


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?

@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

dadgar · 2018-02-27T00:37:14Z

client/client.go

@@ -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),


I would drop to like 8

dadgar · 2018-02-27T00:39:07Z

client/client.go

+			return false
+		}
+		f := second.Networks[i]
+		if e != f {


This comparison is a pointer comparison and not a value comparison?

dadgar · 2018-02-27T00:41:04Z

client/client.go

+
+			hasChanged = false
+			syncTicker.Stop()
+			syncTicker = time.NewTicker(c.retryIntv(nodeUpdateRetryIntv))


Use a timer and reset

dadgar · 2018-02-27T00:41:37Z

client/client_test.go

+	// our linter without explicit disabling.
+	c1 := TestClient(t, func(c *config.Config) {
+		c.Options = map[string]string{
+			driver.ShutdownPeriodicAfter:    "true", // nolint: varcheck


Don't need the nolint

remove impossible test case

dadgar · 2018-02-28T18:28:34Z

client/client.go

+			hasChanged = false
+			timer.Reset(c.retryIntv(nodeUpdateRetryIntv))
+		case <-c.triggerNodeUpdate:
+			hasChanged = true


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))

dadgar · 2018-02-28T18:30:34Z

client/client.go

-				c.logger.Printf("[DEBUG] client: state changed, updating node.")
+		case <-timer.C:
+			if !hasChanged {
+				continue


Right now this code could never fire even if there is a node update

github-actions · 2023-03-11T02:14:12Z

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.
If you have found a problem that seems related to this change, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.

chelseakomlo force-pushed the r-edge-trigger-node-watcher branch from 1313d88 to f4ba5f6 Compare February 16, 2018 16:34

chelseakomlo changed the title ~~[WIP] Edge trigger node updates~~ Edge trigger node updates Feb 16, 2018

chelseakomlo force-pushed the r-edge-trigger-node-watcher branch from f4ba5f6 to 7b0f18b Compare February 16, 2018 20:21

chelseakomlo mentioned this pull request Feb 16, 2018

Client driver health checks for Docker #3856

Merged

dadgar requested changes Feb 22, 2018

View reviewed changes

preetapan reviewed Feb 23, 2018

View reviewed changes

chelseakomlo added 2 commits February 26, 2018 12:36

edge trigger node update

7246c9d

test update config copy trigger

code review fixups

4271123

chelseakomlo force-pushed the r-edge-trigger-node-watcher branch from 8950da5 to 21fa3cd Compare February 26, 2018 17:55

chelseakomlo mentioned this pull request Feb 26, 2018

[Blocked] Use driver health check status to determine scheduling #3905

Closed

chelseakomlo force-pushed the r-edge-trigger-node-watcher branch from 21fa3cd to ac91ee4 Compare February 26, 2018 18:55

improved testing; node networks comparison

cd5be3b

chelseakomlo force-pushed the r-edge-trigger-node-watcher branch from ac91ee4 to cd5be3b Compare February 26, 2018 20:55

use time ticker, update how network resources are compared

44fdb74

dadgar requested changes Feb 27, 2018

View reviewed changes

chelseakomlo added 3 commits February 27, 2018 12:42

add network resources equal method, use time ticker

2969611

remove impossible test case

reset timer after updating node copy

4b6bcf7

fix up linting

d635323

dadgar requested changes Feb 28, 2018

View reviewed changes

simplify retry logic

bf72619

dadgar approved these changes Mar 1, 2018

View reviewed changes

chelseakomlo merged commit a17690c into master Mar 1, 2018

chelseakomlo deleted the r-edge-trigger-node-watcher branch March 1, 2018 20:19

github-actions bot locked as resolved and limited conversation to collaborators Mar 11, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Edge trigger node updates #3873

Edge trigger node updates #3873

chelseakomlo commented Feb 14, 2018 •

edited

Loading

dadgar Feb 22, 2018

dadgar Feb 22, 2018

dadgar Feb 22, 2018

dadgar Feb 22, 2018

dadgar Feb 22, 2018

dadgar Feb 22, 2018

dadgar Feb 22, 2018

preetapan Feb 23, 2018

chelseakomlo Feb 23, 2018

dadgar Feb 27, 2018

dadgar Feb 27, 2018

dadgar Feb 27, 2018

dadgar Feb 27, 2018

dadgar Feb 28, 2018

dadgar Feb 28, 2018

github-actions bot commented Mar 11, 2023


		var nodeHasChanged bool

		for name, new_val := range response.Attributes {

		@@ -1506,6 +1517,9 @@ func (c *Client) watchNodeUpdates(attrHash, metaHash uint64) {

		c.retryRegisterNode()

Edge trigger node updates #3873

Edge trigger node updates #3873

Conversation

chelseakomlo commented Feb 14, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

github-actions bot commented Mar 11, 2023

chelseakomlo commented Feb 14, 2018 •

edited

Loading