github openziti/zrok v2.0.11

one hour ago

FIX: The metrics bridge with a file source could skip usage events, or pass on a partial event that was then discarded, when it read the file while OpenZiti was writing to it. Some usage consequently went unrecorded. The file source now delivers each event once, when its line is complete.

FIX: Usage for a share with more than one connection in a reporting interval was under-reported because each connection's report replaced the one before it. These reports now keep separate points, so totals shown in the console and API will be higher for such shares from the upgrade on. Limits are now enforced against the true totals: an account may reach a warning or limit it did not reach before, and operators may want to review their limit values. Usage recorded before the upgrade is not corrected.

CHANGE: The controller's metrics agent now remembers which share each OpenZiti service belongs to, and which services are not shares at all, instead of asking the database again for every usage event. A busy controller makes far fewer database queries as a result; usage keeps being attributed to the same share, environment and account. A share is remembered for 10 minutes and a service that is not a share for 1 minute, set with metrics.agent.share_cache_lifetime and metrics.agent.not_share_cache_lifetime; metrics.agent.share_cache_size bounds how many are held (100000 by default). A failed database lookup is never remembered. Every 5 minutes, when usage has flowed, the agent logs how many lookups it answered from memory and how many went to the database; metrics.agent.cache_report_interval changes that, and a negative value turns the line off. A negative lifetime turns that kind of remembering off and restores the previous behaviour.

CHANGE: The controller's limits agent now remembers, for a short time, each account's limits and its transfer totals from InfluxDB, and finishes a usage event without a database query or an InfluxDB query when they show there is nothing to do. Warnings and limits are still journaled and applied only from the database and InfluxDB, read at that moment. Enforcement can therefore lag: a change to an account's limit classes or limitless flag can take up to limits.account_cache_lifetime to be noticed, and a crossing of a threshold up to limits.usage_cache_lifetime, 60 seconds each by default. Every 5 minutes, when usage has flowed, the agent logs how many events it finished from memory and how many took the full path; limits.cache_report_interval changes that, and a negative value turns the line off. A negative lifetime turns that cache off; setting both negative restores the previous behaviour.

FIX: The agent can release and clean up an access or a private share that the zrok service has already removed. When the share behind an access was deleted on its own side, the access showed as failed in zrok2 agent status, zrok2 agent release access answered unaccessNotFound, and only restarting the agent cleared it. The agent now confirms through the account overview that the access or share is really gone and finishes the cleanup. Deleting a private share or an access from the agent is also bounded at 30 seconds, as deleting a public share already was.

Don't miss a new zrok release

NewReleases is sending notifications on new releases.