|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This metric attempts to determine how many hosts it's possible to lose
before there's no place to fit its tenants, by finding a "shortest path to failure".
Since finding the actual path is np-hard, this maintainer constructs a
heuristic based on "repeated removals", and finds greedily finds a path
to failure with it.
The Node Alerter also exposes the "overcommittedNodes" metric, counting
how many hosts have children expecting more resources than it can provide.
Finally, this commit adds an obfuscated dump of data from zookeeper,
useful for running tests which require a node repository which reflects
reality.
|