diff --git a/process_queue b/process_queue index c1973e8..85c17b2 100644 --- a/process_queue +++ b/process_queue @@ -55,14 +55,27 @@ class DiskFullPredictor: tuf = now - lts.data[i][0] break else: - # XXX - this is probably a range, so maybe we should use some kind - # of average. It might also be zero, so maybe we have to search for - # the first non-zero value? For now keep it simple. - first_used_bytes = lts.data[0][1] - historic_growth = current_used_bytes / first_used_bytes - future_growth = current_usable_bytes / current_used_bytes - tuf = math.log(future_growth) / math.log(historic_growth) * (now - lts.data[0][0]) - tuf = max(tuf, now - lts.data[0][0]) + # Try always use the minimum of a range. + # We prefer the first datapoint + first_used_bytes = lts.data[0][2] if len(lts.data[0]) >= 4 else lts.data[0][1] + # But if that's not useable we search the whole timeseries for the + # minimum + if first_used_bytes >= current_used_bytes: + first_used_bytes = current_used_bytes + first_i = None + for i in range(len(lts.data)): + used_bytes = lts.data[i][2] if len(lts.data[i]) >= 4 else lts.data[i][1] + if used_bytes < first_used_bytes: + first_used_bytes = used_bytes + first_i = i + else: + first_i = 0 + + if first_i is not None: + historic_growth = current_used_bytes / first_used_bytes + future_growth = current_usable_bytes / current_used_bytes + tuf = math.log(future_growth) / math.log(historic_growth) * (now - lts.data[first_i][0]) + tuf = max(tuf, now - lts.data[first_i][0]) desc = {**lts.description, "measure": "time_until_disk_full", "node": node,