As long as the database is acid restarts should not be a factor. Data integrity is not helped by transactions, you would need error correcting codes for that. Plus the effect on performance is quite notable on all dbs I’ve worked with.
Restarts in a server between dB updates that in a sane world would be txns I meant (e.g update A, crash so don’t update B). Anyway, in postgres they’re pretty cheap in the absence of actual conflict – more expensive if you have actual cinflicts, obvs.
Well it depends how much data integrity is worth to you, and how your system works. Every write in postgres is already a transaction - when you can get away with simple crud stuff, often there’s nothing to do, you have transactionality already. Transaction isolation levels are where db operation costs might change under concurrent conflicting writes but you can tune that by ensuring single-writer-per-partition or whatever in your server logic and it might add a ms or two. OTOH if you have heavy contestation it can be much more expensive. The performance implications are complicated but can certainly kept to a fraction of overall cost depending on your workload!
But adding it to an 80ms operation is. If your operation is 0.5ms it’s either a read on a small table, or maybe a single write – transaction isolation wouldn’t even be relevant there. You’re right that I did mean consistency rather that integrity though, slip of the terminology, but not really worth quibbling over. The point I meant was that I like my data to make sense, a funny quirk of mine.
As long as the database is acid restarts should not be a factor. Data integrity is not helped by transactions, you would need error correcting codes for that. Plus the effect on performance is quite notable on all dbs I’ve worked with.
Restarts in a server between dB updates that in a sane world would be txns I meant (e.g update A, crash so don’t update B). Anyway, in postgres they’re pretty cheap in the absence of actual conflict – more expensive if you have actual cinflicts, obvs.
“Pretty cheap” is very subjective…
Well it depends how much data integrity is worth to you, and how your system works. Every write in postgres is already a transaction - when you can get away with simple crud stuff, often there’s nothing to do, you have transactionality already. Transaction isolation levels are where db operation costs might change under concurrent conflicting writes but you can tune that by ensuring single-writer-per-partition or whatever in your server logic and it might add a ms or two. OTOH if you have heavy contestation it can be much more expensive. The performance implications are complicated but can certainly kept to a fraction of overall cost depending on your workload!
Again, not data integrity (Error correction) but consistency (aCid). Adding two milliseconds to a half a millisecond operation is by no means cheap…
But adding it to an 80ms operation is. If your operation is 0.5ms it’s either a read on a small table, or maybe a single write – transaction isolation wouldn’t even be relevant there. You’re right that I did mean consistency rather that integrity though, slip of the terminology, but not really worth quibbling over. The point I meant was that I like my data to make sense, a funny quirk of mine.
If your single operations take 80ms either it’s a toy app or someone didn’t do their job (unoptimized queries, wrong technology, wrong modeling, etc).