timestamps: formatting and timezones

[Zooko]

Please use ISO-8601, except with "_" instead of "T" for the separating character for all timestamps. Please include the ISO-8601 unambiguous timezone indicator (which is either "Z" or a number of hours offset from UTC -- I don't know what the standard is for non-integral offsets, but see below).

Next, please consider using UTC times for all timestamps instead of localtimes. Foolscap is system for remote computation, and the remote end may or may not be in the same timezone as the local end, which may or may not be in the same timezone as the programmer who is looking at it. A nice, consistent simple way to deal with such things is just not worry as much about how sunny the sky was, and who was awake, in the land where the server lives and just use UTC for everything.

(More seriously, the trade-off is cognitive difficulty of correlating events on a server from human-rhythm events that take place in the same timezone, such as meals and sleeping, vs. the cognitive difficulty of correlating events on a server with events on other computers. In my experience working in distributed systems, I'm more frequently interested in the latter correlations.)

Here is the related discussion from the Tahoe trac:

• ​http://allmydata.org/trac/tahoe/ticket/326 # use consistent time stamps in logging

Here are some Python implementations of ISO and/or UTC timestamps:

• ​http://allmydata.org/trac/tahoe/browser/src/allmydata/node.py?rev=2731#L30
• ​http://allmydata.org/trac/tahoe/browser/src/allmydata/util/time_format.py?rev=2424#L12
• ​http://allmydata.org/trac/pyutil/browser/pyutil/pyutil/time_format.py?rev=93

[Zooko]

Carrying over the conversation from tahoe #326. Brian wrote: "I am often frustrated when looking at a log file and trying to figure out whether this is the operation that I just did ten minutes ago, or if it is from last night, and the eight-hour PST offset is usually almost exactly the difference between these two.".

I've already commented about a difference between Brian's and my experience: the "which timezone is your computer in?" issue. It is a business goal of allmydata.com to have servers in more timezones in the near future.

The other thing that I just realized that differs between Brian's and my experiences is that Brian is often investigating an operation that he did just ten minutes ago (or that he did last night), while I have often been investigating something that happened on someone else's schedule -- either it was one of our customers, or it was a purely automated event with no correlation with human activity.

I anticipate that for Tahoe grids that get used in the future, we will less frequently want to correlate logs with events that we remember from our recent personal lives.

[Zooko]

I really should have made this two tickets. The first part is urgent for me, and I don't think there is any disagreement about it: make timestamps have an unambiguous format, date, and timezone indicator.

[Brian Warner]

minor summary reformatting

[Brian Warner]

Haven't we pretty much wrapped this one up? I remember a bunch of time-formatting changes (in flogtool, mostly) that I think came to some sort of compromise.

[Zooko]

The following related tickets are already closed as fixed: #112 (timestamps of incident files -- ISO-8601'ish), #111 (Incident filename timestamps: TZ, UTC, ISO-8601 please), #100 (timestamps: use fully qualified, unambiguous, standard format). I think those address all of the issues in this ticket.