Troubleshooting CCF

This page contains troubleshooting tips for CCF.

Tips for interacting with CCF to diagnose issues

Note

In the examples below this documentation uses example-ccf-domain.com as an example CCF domain, you will need to replace that with your own CCF domain when using these commands. You will also need to add authentication parameters such as --cacert to the curl commands, see Issuing commands for an example.

Note

CCF may be deployed with a load balancer which may cache the node which last responded to a query from an IP address. Until the cache clears, the load balancer will direct any subsequent queries from that IP address to the same node. As an example, if the cache clears after one minute, then in order to get a response from a different node, an operator must wait one minute between queries.

Below are descriptions of CLI commands and how they are useful for diagnosing CCF issues:

“What node is handling my requests?”

curl https://example-ccf-domain.com/node/network/nodes/self -i

This is useful to identify which node is handling queries. The node ID can be found in the location header as shown in the example command output below:

HTTP/1.1 308 Permanent Redirect
content-length: 0
location: https://example-ccf-domain/node/network/nodes/<Node ID>

“What CCF version is running?”

curl https://example-ccf-domain.com/node/version

This is useful to confirm the version that is running.

“What nodes are part of the current network?”

curl https://example-ccf-domain.com/node/network/nodes

This will show information for all nodes in the network. In a healthy network all nodes will show “status”: “Trusted”, and one node only will show “primary” = true. This is the healthy state of the network. Around upgrades/restarts/migrations nodes will transition through unhealthy states temporarily. If the network remains in an unhealthly state for a long time, this indicates there is an issue.

You can obtain this information for a single node by querying the GET /node/network/nodes/{node_id} endpoint, where {node id} can be obtained from the GET /node/network/nodes/self endpoint described above. Take note of the node_data field in the response which contains useful correlation IDs.

“Is the network in the middle of a reconfiguration?”

curl https://example-ccf-domain.com/node/consensus

This has a few bits of data that might help us diagnose a partitioned/faulty network. In particular, most of the time there should be a single entry in the configs list. During an upgrade/restart/migration, there may be multiple values. If multiple values persist for a long time, it suggests something went wrong during the reconfiguration.

“Is the CCF network stable?”

curl https://example-ccf-domain.com/node/commit

This is a good endpoint to query to check if the CCF service is reachable. Additionally, a large and increasing difference between the View in the Transaction ID in this response, and the current_view from the GET /node/consensus response, indicates a partitioned node. For example, if the response from GET /node/commit shows the View is 15, and the response from GET /node/consensus states the current view is 78967 and that number is constantly increasing, then this indicates the node is unable to make consensus progress, which likely indicates it is unable to contact other nodes.

Tip

See tests/infra/health_watcher.py for a detailed technical example of how the health of the network can be monitored.

Node Output

By default node output is written to stdout and to stderr and can be handled accordingly.

There is an option to generate machine-readable logs for monitoring. To enable this, set the logging.format configuration entry to "Json". The generated logs will be in JSON format as displayed below:

{
    "e_ts": "2019-09-02T14:47:24.589386Z",
    "file": "../src/consensus/aft/raft.h",
    "h_ts": "2019-09-02T14:47:24.589384Z",
    "level": "info",
    "msg": "Deserialising signature at 24\n",
    "number": 651
}

  • e_ts is the ISO 8601 UTC timestamp of the log if logged inside the enclave (field will be missing if line was logged on the host side)

  • h_ts is the ISO 8601 UTC timestamp of the log when logged on the host side

  • file is the file the log originated from

  • number is the line number in the file the log originated from

  • level is the level of the log message [info, debug, trace, fail, fatal]

  • msg is the log message

See this page for steps to add application-specific logging, which will have an additional tag field set to app.

Error Codes

StartupSeqnoIsOld

Returned when a node tries to join a network with too old a snapshot, or no snapshot at all. See this page for more information.

Resolution: This can be resolved by trying to join again with a fresh snapshot. The seqno of the snapshot a node started from is available as startup_seqno in GET /node/state.