test-1.6-ci-1

Disable tests in ci

They appear to block the CI.

v1.8.222-criteo6

Fix ci (#17)

It appears the default Travis script for sbt was conflicting with the version used in Marathon.
On this occasion, we chose to migrate to Github Actions because it allowed the build to pass, including tests.
Not removing Travis related files since they are tracked upstream and can contain modification reflecting impacting changes in the future.

v1.8.222-criteo5

Anti-snowball: Prevent Marathon from killing too many unhealthy insta…

v1.8.222-criteo3: Fix Marathon backoff

This includes the following original commits:
f419cee1f Titouan Chary Decrease backoff delay when application are Running
0959bdc38 Titouan Chary Fix marathon backoff issue when backoff are under 1.1

v1.8.222-criteo2: Add support for network bandwidth to Marathon

This change contains the following original commits:
0f8f2b6f5 Grégoire Seux Make sure networkBandwidth can be specified in all scenarii
8be4d7f2e Titouan Chary Fix network bandwidth tests for Pods
9dbb4d6dc Clement Michaud Add network bandwidth in RAML files
ef84d5a65 Clement Michaud Fix and add tests to support network bandwidth
0a59fe067 Clement Michaud Add support for network bandwidth to Marathon
ee1c7dbe3 Titouan Chary Fix tests for networkBandwidth naming

v1.6.564-criteo3: Properly detect travis branch

Change-Id: I49a6e54ef9f8ad0fd0e227f53a3d60d1b822a849

untagged-a8db1f6776e478b9d299: Publish tag on every commit in criteo/* branch (#16)

Change-Id: I215d6a14ce6e9dbafbff4d58f2d6d2346e090860

Latest Anti-snowball patch

Anti snowball round 4: let's make it opt-in (somehow) (#14)

Make anti-snowball opt-in

Allow all application instances to be killed if anti-snowball has not been enabled (upgradeStrategy.minimumHealthCapacity set to default value 1).

v1.6.564-criteo1: Fix anti-snowball (#12)

Fix anti-snowball

* Count "ready" instances instead of active ones

Previous implementation was based on active instance count ("active"
including not only running instances but created, stating, and even
killing ones, see core.condition.Condition for details), which naturally
lead to unhealthy/protected instances being killed too soon (as soon as
replacement instances are spawned in fact).
We consider an instance ready if its health check has succeeded at least
once.

* Remove synchronous blocks

This reverts the handleHealthResult to the pre-anti-snowball version,
thus removing any synchronous call.
We also revert the synchronous part of checkConsecutiveFailures.
And now, the checkConsecutiveFailures is executed through an actor
message, guaranteeing that only 1 kill order will be computed at once.

v1.6.564-criteo1-snowball1

Limit killingInFlight growth