some light questions

[cellio]

I learned today that there is a full-service gas station on my way to/from work. I didn't know we had any of those locally. It's been years (probably decades); what is the conventional tip?

As I pulled up to an intersction (all-way stop), someone from the cross street was backing through the intersection. After backing into the space in front of my car, he immediately popped into drive and went through the intersection. Whose turn was that, the cross-street or mine? :-)

I have occasionally noticed (because of tracking/RSS feeds or because I viewed the journals directly) posts to LJ that did not show up on my friends page. Is this happening to anyone else? I haven't detected a pattern yet.

Why does Hebrew have two words for "open" that differ only (apparently) in what objects they take? It's peh-kuf-chet when talking about eyes and ears, and peh-taf-chet for anything else.

Comments

[cellio]

you're talking about unlocked posts that you can see if you visit the person's LJ directly, but that don't show up on your friends page, right?

Correct.

I have noticed that there are two people on my f-list who seem particularly prone to that phenomenon, but like you I haven't been able to find any other pattern to it.

That's possibly the beginnings of a pattern, at least. If the affected users are all on the same flaky cluster, for example, that could do it. I haven't seen enough instances yet to have any user-specific patterns -- it's been one here, one there, and I haven't noticed it being the same people (yet). I'll be on the lookout for that now.

[jducoeur]

Having been digging into LJ's architecture lately, it wouldn't surprise me at all if it's a cross-cluster problem. From at least the 1000-foot view, I don't seen anything that really maintains referential integrity between clusters (indeed, it would be pretty hard to do). So it's easy to imagine some flakiness causing, eg, an update message getting lost when crossing cluster lines. I assume that they guard against that, but catching every possible case in a formal and guaranteed way is pretty hard.

(As I contemplate trying to scale CommYou up, this is one of my major considerations. My CIO is of the opinion that I should just throw money at the problem when we get to that point, and spend a million bucks on an Oracle rack, rather than try to play fancy multi-cluster games. Terrifying idea, but he may have a point. If the monetization strategy works at all, we should have enough income to pull it off when it becomes a severe issue...)