Files
zeekay 833094dc48 status: the module is hanzo.ai/status
The import path a reader types is the one the org owns, rather than the forge
that happens to host the bytes. hanzo.ai serves the go-import meta and points it
at the same repository, so where the code lives does not change.

Rewritten by role. A bare repository URL — a clone target, a badge, a build
context — still names where the code is and stays on github.com.
2026-08-23 19:48:00 -07:00

109 lines
4.1 KiB
Go

package memory
import (
"hanzo.ai/status/config/endpoint"
"hanzo.ai/status/config/suite"
"hanzo.ai/status/storage/store/common/paging"
)
// ShallowCopyEndpointStatus returns a shallow copy of a Status with only the results
// within the range defined by the page and pageSize parameters
func ShallowCopyEndpointStatus(ss *endpoint.Status, params *paging.EndpointStatusParams) *endpoint.Status {
shallowCopy := &endpoint.Status{
Name: ss.Name,
Group: ss.Group,
Key: ss.Key,
Uptime: endpoint.NewUptime(),
}
if params == nil || (params.ResultsPage == 0 && params.ResultsPageSize == 0 && params.EventsPage == 0 && params.EventsPageSize == 0) {
shallowCopy.Results = ss.Results
shallowCopy.Events = ss.Events
} else {
numberOfResults := len(ss.Results)
resultsStart, resultsEnd := getStartAndEndIndex(numberOfResults, params.ResultsPage, params.ResultsPageSize)
if resultsStart < 0 || resultsEnd < 0 {
shallowCopy.Results = []*endpoint.Result{}
} else {
shallowCopy.Results = ss.Results[resultsStart:resultsEnd]
}
numberOfEvents := len(ss.Events)
eventsStart, eventsEnd := getStartAndEndIndex(numberOfEvents, params.EventsPage, params.EventsPageSize)
if eventsStart < 0 || eventsEnd < 0 {
shallowCopy.Events = []*endpoint.Event{}
} else {
shallowCopy.Events = ss.Events[eventsStart:eventsEnd]
}
}
return shallowCopy
}
// ShallowCopySuiteStatus returns a shallow copy of a suite Status with only the results
// within the range defined by the page and pageSize parameters
func ShallowCopySuiteStatus(ss *suite.Status, params *paging.SuiteStatusParams) *suite.Status {
shallowCopy := &suite.Status{
Name: ss.Name,
Group: ss.Group,
Key: ss.Key,
}
if params == nil || (params.Page == 0 && params.PageSize == 0) {
shallowCopy.Results = ss.Results
} else {
numberOfResults := len(ss.Results)
resultsStart, resultsEnd := getStartAndEndIndex(numberOfResults, params.Page, params.PageSize)
if resultsStart < 0 || resultsEnd < 0 {
shallowCopy.Results = []*suite.Result{}
} else {
shallowCopy.Results = ss.Results[resultsStart:resultsEnd]
}
}
return shallowCopy
}
func getStartAndEndIndex(numberOfResults int, page, pageSize int) (int, int) {
if page < 1 || pageSize < 0 {
return -1, -1
}
start := numberOfResults - (page * pageSize)
end := numberOfResults - ((page - 1) * pageSize)
if start > numberOfResults {
start = -1
} else if start < 0 {
start = 0
}
if end > numberOfResults {
end = numberOfResults
}
return start, end
}
// AddResult adds a Result to Status.Results and makes sure that there are
// no more than MaximumNumberOfResults results in the Results slice
func AddResult(ss *endpoint.Status, result *endpoint.Result, maximumNumberOfResults, maximumNumberOfEvents int) {
if ss == nil {
return
}
if len(ss.Results) > 0 {
// Check if there's any change since the last result
if ss.Results[len(ss.Results)-1].Success != result.Success {
ss.Events = append(ss.Events, endpoint.NewEventFromResult(result))
if len(ss.Events) > maximumNumberOfEvents {
// Doing ss.Events[1:] would usually be sufficient, but in the case where for some reason, the slice has
// more than one extra element, we can get rid of all of them at once and thus returning the slice to a
// length of MaximumNumberOfEvents by using ss.Events[len(ss.Events)-MaximumNumberOfEvents:] instead
ss.Events = ss.Events[len(ss.Events)-maximumNumberOfEvents:]
}
}
} else {
// This is the first result, so we need to add the first healthy/unhealthy event
ss.Events = append(ss.Events, endpoint.NewEventFromResult(result))
}
ss.Results = append(ss.Results, result)
if len(ss.Results) > maximumNumberOfResults {
// Doing ss.Results[1:] would usually be sufficient, but in the case where for some reason, the slice has more
// than one extra element, we can get rid of all of them at once and thus returning the slice to a length of
// MaximumNumberOfResults by using ss.Results[len(ss.Results)-MaximumNumberOfResults:] instead
ss.Results = ss.Results[len(ss.Results)-maximumNumberOfResults:]
}
processUptimeAfterResult(ss.Uptime, result)
}