Console #
The console log is (by default) very brief.
It outputs when a job’s input history changed (so you know why a job was rerun).
It outputs a short stack trace when jobs fail, and the name of the error log file (see below) which contains more details.
And it shows a counter for
- T: Total jobs
- D: Done jobs
- R: Running jobs
- W: Waiting jobs
- F: Failed jobs
Log file(s) #
The log file (.ppg/logs/latest.messages) contains more info.
Compared to the console log it omits the counters, but logs every job start. For jobs that took more than one second (or if the log level is <= log_debug), every job finish is also logged.
It’s messages contain references into the ppg2 source, these are abstracted away into latest.lookup to denoise the log.
You can increase the amount of logging by passing a lower low level to ppg2.new()
Error log files #
For each failed job, an error log in .ppg/errors/latest/<job_number>_exception.txt.
It contains the full stack trace (including locals!), the message of the exception, and - if available - stdout and stderr.
Log rotation #
By default the log files and error logs of the last few ppg2 runs are kept. Technically, we store each log under the run date, and symlink ’latest’ to the latest one.
Runtimes: #
Every time a job takes more than a second, we log it’s runtime at the end of the run to .ppg/logs/runtimes.tsv.
History #
The information necessary to decide wether jobs are invalidate is stored in .ppg/history/ppg_history.2.gz. It’s a gziped json, key -> recorded job output.
The keys follow the following format:
- “A” -> The last output of job “A”
- “A!!!” -> The job_ids of the inputs of job “A” (to detect when an input is added or removed)
- “A!!!B” the last output of job “A” that was used as input for job “B”.