checking in changes in verify-loader script for log loss simulation #17
Conversation
verify-loader
Outdated
| if not line.startswith("loader seq - "): | ||
| #find the log header: loader seq - | ||
| #for container logs the line will not start with header - instead there's a timestamp; if header not present ignore this line | ||
| if "loader seq - " not in line: |
There was a problem hiding this comment.
If the log lines read by verify-loader have a fixed prefix added to them, then something else should process that header away before sending the log lines to the verify-loader.
I don't think we should try to endow verify-loader with an understanding of how to pull out the log lines.
verify-loader
Outdated
| #check that line read has constituent parts after header (<header> - <uuid> - <seq_num> - <payload>) | ||
| try: | ||
| _, invocid, seqval, payload = line.split('-', 4) | ||
| indx = line.find('loader seq -') |
There was a problem hiding this comment.
Same as above, we don't want to add support for prefixed data of the log lines generated by the loader. If there is a prefix, let's have another tool strip it before sending the data to verify-loader.
verify-loader
Outdated
| if prev is not None: | ||
| # Bad record encountered, flag it | ||
| print("%s: %d %d <-" % (invocid, seq, prev)) | ||
| loss_count += (seq-prev) |
There was a problem hiding this comment.
So if seq is less than prev this will be a negative value. I don't think we want to account for loss that way.
Instead we might want to consider two conditions: seq > prev + 1 and seq <= prev.
The add the distance between seq and prev makes sense on the first condition. But for the second condition we'll want to keep track of contiguous ranges, expanding the range as the sequence grows, creating a new sequence when a gap is encountered, and looking for duplicates by seeing if the new seq is in any known ranges.
verify-loader
Outdated
| (total_count / (now - start)), | ||
| (ignored_bytes / MB) / (now - start), | ||
| (ignored_count / (now - start)))) | ||
| print("interval stats:: total bytes: %d, total lines: %d, ignored: %d" % (report_bytes, report_count, report_ignored_count)) |
There was a problem hiding this comment.
Why not add this data to the previous print statement?
verify-loader
Outdated
| (ignored_bytes / MB) / (now - start), | ||
| (ignored_count / (now - start)))) | ||
| print("total bytes: %d, total lines: %d, ignored lines: %d, lost(out-ofseq) lines: %d" % (total_bytes, total_count, ignored_count, loss_count)) | ||
| print("overall loss percentage = %.3f" %(loss_count*100.0/total_count)) |
There was a problem hiding this comment.
Why not add this data to the previous print statement?
|
@atrimandal: PR needs rebase. Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
3 participants
Script changes for simulating log loss