During a meeting it was brought up to me that the OpenShift/Kubernetes logging strategy isn’t very concise. Though looking into this I wanted to put some context around the technology. “How does OpenShift capture logs?” “What is captured and logged?” “What is my recommendations for using the logging system?”
EFK stands for Elastic Search (E), Fluentd (F), and Kibana (K). This is a modification on the traditional ELK stack that has become popular in recent years for log aggregation, collection and sorting. Kibana acts as the user interface for the collected logs. Elastic Search is the search and analytics engine. Fluentd is a unified logging system with hundreds (500+ as of the time if this writing ) of plugins.
What is captured
Looking thought the Kubernetes documentation, it’s made a bit more clear what is captured where and how applications logs are managed from the container level. In the section titled ‘Logging at the node level’  it is explained that “Everything a containerized application writes to stdout and stderr is handled and redirected somewhere by a container engine. For example, the Docker container engine redirects those two streams to a logging driver, which is configured in Kubernetes to write to a file in json format.” It is said in OpenShift documentation “Fluentd reads from /var/log/messages and /var/log/containers/.log for system logs and container logs, respectively. You can instead use the systemd journal as the log source. There are three deployer configuration parameters available in the deployer ConfigMap.” . For additional information and resources on Fluentd, I strongly recommend watching the ‘OpenShift Commons’ videos from May 17, 2017 .
Cluster wide -vs- Project logging
This is not a simple question to answer. I’m drawing upon my experience working with other clustered technologies and customers that have implemented OpenShift. My recommendation is to do what’s right for your environment. I know… not very useful. Hopefully my heuristics will lead you to your answer.
More often than not your company, organization, group and team have their structure. I’ve worked with companies and agencies that have had every sort of organically grown business structure. Some extremely independent, some centralized, and some ignorant to the structure entirely. We have to consider how you do business today and what will actually work and how we can fit into that system. What are the security requirements? What are the data retention rates? What is the disaster recovery strategy?
If I were to propose that every project have their own EFK stack to manage only their logs. A customer running 100+ project will have a LOT of redundancy and the overhead for a security team to manage and track logs could be prohibitory expensive/complicated. How does a security team monitor the creation of new projects, validated their access and ultimate ensure the security and compliance of the systems?
If I proposed one giant company-wide EFK stack it would lighten the burden for some but could cause data management and growth complications. Our security team is happy, because they have one log on to one server to see all the system and application logs being generated by the containers and applications. Let me assume for a minute a non-common use case for OpenShift, batch processing. I want to use this platform to ETL function on a file I have stored out in S3. That project or job that live and die on my whim, might introduce long stale data into my logging system and tool chain. The point of a job is to run and be gone, so I might not care about the details.
While working as a US Army consultant in 2011, we were implementing Splunk. Working though the data ingest rates and figuring out what was good and stale data was complicated and we had fairly static workloads. Working though all the requirements will likely guide you the right direction. I suggest pruning what is important and measuring them often, high signal to noise ratio. This typically means smaller units or project based logging. It becomes quite daunting to measure every job, application and container in your environment on an ongoing basis. Off load that responsibility to the application and project owners.
Since I mentioned Splunk, I thought it is important to include the following section as well. ‘Configuring Fluentd to Send Logs to an External Log Aggregator’. You can configure Fluentd to send a copy of its logs to an external log aggregator, and not the default Elasticsearch, using the secure-forward plug-in. From there, you can further process log records after the locally hosted Fluentd has processed them .
Fluentd Plug ins list
Logging at the node level
Aggregate Logging – OpenShift Docs
Fluentd – OpenShift Commons Briefing