DepC Documentation¶
DepC (Dependency Checker) is a Qos Measurement & Dependency Graph Platform created by OVH. We use it to store and request our CMDB and to compute the QoS of our infrastructure, including our customers.
Overview¶
DepC provides a CMDB component to store and request related nodes into a graph
structure. These nodes can be anything : servers, products, customers or even
web services. A typical use case would be to create some customers
nodes,
related to their products
. We can also link these products
to
servers
. that way it would be very easy to display the impacted customers
when a problem occurs on a server.
Once DepC knows your dependency graph, it then becomes possible to compute a
QoS for every nodes. By taking up our previous example, Depc can compute the
QoS of our customers
, following the QoS of their servers
.
Note
QoS is not the same as SLA : please read this guide to understand the difference between them. Notions about SLO and Indicators (used by DepC to compute the QoS) are also explained in it.
Principles¶
- Graph Dependency : We use the Neo4j database to manage the nodes and their relationships (we already handle several million of nodes in our OVH internal instance). DepC provides some API calls and WebUI to easily request a node and its relationships.
- QoS Computation : DepC can compute a QoS for all nodes in the Graph, but
of course we have to explain how we do it. We created different methods for that : some nodes will use the raw data stored in TimeSeries
databases (for example a
server
node having a probe), other nodes will use their parents QoS to compute their own one. Finally the QoS can be displayed in Grafana or in the DepC WebUI (this last one allows some cool features, like display the worst nodes or even the root cause of a bad QoS). - Scalability : Apache Kafka is used to receive payloads and forward them to Neo4j, adding the wanted scalabily and high-availability management. Then Apache Airflow is responsible to automate the QoS computation, so DepC can benefit from its executors feature (Celery, Kubernetes..) to scale horizontally.