Additional Heartbeat Reload Tests #8228
Conversation
heartbeat/monitors/pluginconf.go
Outdated
| return fmt.Sprintf("Monitor not loaded, plugin is disabled") | ||
| } | ||
|
|
||
| type MonitorPluginInfo struct { |
There was a problem hiding this comment.
exported type MonitorPluginInfo should have comment or be unexported
heartbeat/monitors/pluginconf.go
Outdated
| "github.com/elastic/beats/libbeat/common" | ||
| ) | ||
|
|
||
| type PluginDisabledError struct{} |
There was a problem hiding this comment.
exported type PluginDisabledError should have comment or be unexported
heartbeat/monitors/monitor.go
Outdated
| } | ||
| } | ||
|
|
||
| func (m *Monitor) Stop() { |
There was a problem hiding this comment.
exported method Monitor.Stop should have comment or be unexported
heartbeat/monitors/monitor.go
Outdated
| return nil | ||
| } | ||
|
|
||
| func (m *Monitor) Start() { |
There was a problem hiding this comment.
exported method Monitor.Start should have comment or be unexported
andrewvc
force-pushed
the
hbreload-test
branch
3 times, most recently
from
42df612 to
f766544
Compare
andrewvc
changed the title
andrewvc
added
review
and removed
in progress
|
This is now ready for review |
heartbeat/monitors/monitor_test.go
Outdated
| require.Equal(t, 1, len(pipelineConnector.clients)) | ||
| pcClient := pipelineConnector.clients[0] | ||
|
|
||
| time.Sleep(time.Duration(100 * time.Millisecond)) |
andrewvc
force-pushed
the
hbreload-test
branch
from
c4a805c to
35a9170
Compare
|
@ruflin fixed and improved in 35a9170 The issue was that the mock impl was doing a copy into a zero size slice with the right capacity, instead of an empty slice of the right capacity. I also improved the timeout code to exit early by spinning on the condition. Wouldn't mind some review on this, my understanding is that the go scheduler is not safe to spin in this situation and still make forward-progress without explicit sleeps in it, even on a single core system. Is that correct? I added microsecond sleeps in the loop to make sure we yield. Looking at golang/go#10958 seems to confirm that aspect of the go scheduler behavior. |
|
It seems on Jenkins Mac build this causes a problem in libbeat: https://beats-ci.elastic.co/job/elastic+beats+pull-request+multijob-darwin/483/beat=libbeat,label=macosx/console I can't really see how this is related to the interface change in libbeat TBH. Perhaps it's flaky but it's not one I have seen before. |
andrewvc
force-pushed
the
hbreload-test
branch
from
35a9170 to
edc168c
Compare
* Adds a mocked test of the Monitor type * Minimizes the required mockable area by introducing a PipelineConnector type that is easier to mock than a Pipeline. * Introduces a number of useful mocks for future tests potentially. (cherry picked from commit 0417f98)
* Adds a mocked test of the Monitor type * Minimizes the required mockable area by introducing a PipelineConnector type that is easier to mock than a Pipeline. * Introduces a number of useful mocks for future tests potentially. (cherry picked from commit 0417f98)
This PR attempts to improve the testing and testability of heartbeat in the following ways:
MonitortypePipelineConnectortype that is easier to mock than aPipeline.