High Availability Config
For high-traffic websites, we recommend using a high-availability configuration for Fluentd.
See also The Patterns of Distributed Logging and Containers slide for understanding pros/cons of deployment patterns.
Message Delivery Semantics
Fluentd is designed primarily for event-log delivery systems.
In such systems, several delivery guarantees are possible:
At most once: Messages are immediately transferred. If the
transfer succeeds, the message is never sent out again. However,
many failure scenarios can cause lost messages (e.g. no more write
capacity).
At least once: Each message is delivered at least once. In failure
cases, messages may be delivered twice.
Exactly once: Each message is delivered once and only once. This
is the most desirable.
If the system "can't lose a single event" and must also transfer "exactly once", then the system must stop ingesting events when it runs out of write capacity. The proper approach would be to use synchronous logging and return errors when the event cannot be accepted.
That's why Fluentd provides "at most once" and "at least once" transfers. To collect massive amounts of data without impacting application performance, a data logger must transfer data asynchronously. This improves performance at the cost of potential delivery failures.
However, most failure scenarios are preventable. The following sections describe how to set up Fluentd's topology for high-availability.
Network Topology
To configure Fluentd for high-availability, we assume that your network consists of log forwarders and log aggregators.
'Log forwarders' are typically installed on every node to receive local events. Once an event is received, they forward it to the 'log aggregators' through the network. For log forwarders, fluent-bit is also good candidate for light-weight processing.
'Log aggregators' are daemons that continuously receive events from the log forwarders. They buffer the events and periodically upload the data into the cloud.
Fluentd can act as either a log forwarder or a log aggregator, depending on its configuration. The next sections describe the respective setups. We assume that the active log aggregator has an IP 192.168.0.1 and the backup has IP 192.168.0.2.
Log Forwarder Configuration
Use the following configuration to configure log forwarders to transfer logs to log aggregators:
When the active aggregator (192.168.0.1) dies, the logs will instead be sent to the backup aggregator (192.168.0.2). If both servers die, the logs are buffered on-disk at the corresponding forwarder nodes.
Log Aggregator Configuration
Use the following configuration for log aggregators to configure the input source for the log transfer as TCP:
The incoming logs are buffered, then periodically uploaded to the cloud. If the upload fails, the logs are stored on the local disk until the retransmission succeeds.
Failure Case Scenarios
Forwarder Failure
When a log forwarder receives events from applications, the events are first written into a disk buffer (specified by <buffer>'s path). After every flush_interval, the buffered data is forwarded to aggregators.
This process is inherently robust against data loss. If a log forwarder's fluentd process dies, then on its restart, the buffered data is properly transferred to its aggregator. If the network between forwarders and aggregators breaks, the data transfer is automatically retried.
However, possible message loss scenarios do exist:
The process dies immediately after receiving the events, but before writing
them into the buffer.
The forwarder's disk is broken, and the file buffer is lost.
Aggregator Failure
When log aggregators receive events from log forwarders, the events are first written into a disk buffer (specified by <buffer>'s path). After every flush_interval, the buffered data is uploaded to the cloud.
This process is inherently robust against data loss. If a log aggregator's fluentd process dies, then on its restart, the data from the log forwarder is properly retransferred. If the network between aggregators and the cloud breaks, the data transfer is automatically retried.
However, possible message loss scenarios do exist:
The process dies immediately after receiving the events, but before writing
them into the buffer.
The aggregator's disk is broken, and the file buffer is lost.
Troubleshooting
"no nodes are available"
Please make sure that you can communicate with port 24224 using not only TCP, but also UDP. These commands will be useful for checking the network configuration:
Please note that there is one known issue where VMware will occasionally lose small UDP packages used for the heartbeat.
If this article is incorrect or outdated, or omits critical information, please let us know. Fluentd is an open-source project under Cloud Native Computing Foundation (CNCF). All components are available under the Apache 2 License.
Last updated
