func (p *Prospector) startHarvester(info *prospectorInfo, fileconfig *core.FileConfig) {
var offset int64
if p.from_beginning {
offset = 0
} else {
offset = info.identity.Stat().Size()
}
// Send a new file event to allow registrar to begin persisting for this harvester
p.registrar_spool.Add(registrar.NewDiscoverEvent(info, info.file, offset, info.identity.Stat()))
p.startHarvesterWithOffset(info, fileconfig, offset)
}
func (p *Prospector) scan(path string, config *core.FileConfig) {
// Evaluate the path as a wildcards/shell glob
matches, err := filepath.Glob(path)
if err != nil {
log.Error("glob(%s) failed: %v", path, err)
return
}
// Check any matched files to see if we need to start a harvester
for _, file := range matches {
// Check the current info against our index
info, is_known := p.prospectorindex[file]
// Stat the file, following any symlinks
// TODO: Low priority. Trigger loadFileId here for Windows instead of
// waiting for Harvester or Registrar to do it
fileinfo, err := os.Stat(file)
if err == nil {
if fileinfo.IsDir() {
err = newProspectorSkipError("Directory")
}
}
if err != nil {
// Do we know this entry?
if is_known {
if info.status != Status_Invalid {
// The current entry is not an error, orphan it so we can log one
info.orphaned = Orphaned_Maybe
} else if info.err.Error() == err.Error() {
// The same error occurred - don't log it again
info.update(nil, p.iteration)
continue
}
}
// This is a new error
info = newProspectorInfoInvalid(file, err)
info.update(nil, p.iteration)
// Print a friendly log message
if _, ok := err.(*ProspectorSkipError); ok {
log.Info("Skipping %s: %s", file, err)
} else {
log.Error("Error prospecting %s: %s", file, err)
}
p.prospectors[info] = info
p.prospectorindex[file] = info
continue
} else if is_known && info.status == Status_Invalid {
// We have an error stub and we've just successfully got fileinfo
// Mark is_known so we treat as a new file still
is_known = false
}
// Conditions for starting a new harvester:
// - file path hasn't been seen before
// - the file's inode or device changed
if !is_known {
// Check for dead time, but only if the file modification time is before the last scan started
// This ensures we don't skip genuine creations with dead times less than 10s
if previous, previousinfo := p.lookupFileIds(file, fileinfo); previous != "" {
// Symlinks could mean we see the same file twice - skip if we have
if previousinfo == nil {
p.flagDuplicateError(file, info)
continue
}
// This file was simply renamed (known inode+dev) - link the same harvester channel as the old file
log.Info("File rename was detected: %s -> %s", previous, file)
info = previousinfo
info.file = file
p.registrar_spool.Add(registrar.NewRenamedEvent(info, file))
} else {
// This is a new entry
info = newProspectorInfoFromFileInfo(file, fileinfo)
if fileinfo.ModTime().Before(p.lastscan) && time.Since(fileinfo.ModTime()) > config.DeadTime {
// Old file, skip it, but push offset of file size so we start from the end if this file changes and needs picking up
log.Info("Skipping file (older than dead time of %v): %s", config.DeadTime, file)
// Store the offset that we should resume from if we notice a modification
info.finish_offset = fileinfo.Size()
p.registrar_spool.Add(registrar.NewDiscoverEvent(info, file, fileinfo.Size(), fileinfo))
} else {
// Process new file
log.Info("Launching harvester on new file: %s", file)
p.startHarvester(info, config)
}
}
// Store the new entry
p.prospectors[info] = info
} else {
if !info.identity.SameAs(fileinfo) {
// Keep the old file in case we find it again shortly
info.orphaned = Orphaned_Maybe
if previous, previousinfo := p.lookupFileIds(file, fileinfo); previous != "" {
// Symlinks could mean we see the same file twice - skip if we have
if previousinfo == nil {
p.flagDuplicateError(file, nil)
continue
}
// This file was renamed from another file we know - link the same harvester channel as the old file
log.Info("File rename was detected: %s -> %s", previous, file)
info = previousinfo
info.file = file
p.registrar_spool.Add(registrar.NewRenamedEvent(info, file))
} else {
// File is not the same file we saw previously, it must have rotated and is a new file
log.Info("Launching harvester on rotated file: %s", file)
// Forget about the previous harvester and let it continue on the old file - so start a new channel to use with the new harvester
info = newProspectorInfoFromFileInfo(file, fileinfo)
// Process new file
p.startHarvester(info, config)
}
// Store it
p.prospectors[info] = info
}
}
// Resume stopped harvesters
resume := !info.isRunning()
if resume {
if info.status == Status_Resume {
// This is a filestate that was saved, resume the harvester
log.Info("Resuming harvester on a previously harvested file: %s", file)
} else if info.status == Status_Failed {
// Last attempt we failed to start, try again
log.Info("Attempting to restart failed harvester: %s", file)
} else if info.identity.Stat().ModTime() != fileinfo.ModTime() {
// Resume harvesting of an old file we've stopped harvesting from
log.Info("Resuming harvester on an old file that was just modified: %s", file)
} else {
resume = false
}
}
info.update(fileinfo, p.iteration)
if resume {
p.startHarvesterWithOffset(info, config, info.finish_offset)
}
p.prospectorindex[file] = info
} // for each file matched by the glob
}