On Sunday, January 7, 2018 at 2:45:30 PM UTC-5, Massimo Di Pierro wrote: > > I am curious whether anybody has actually tried this. > > I am concerned about a few things that can go wrong: > 1) even GET request may write to the DB (maybe write session in DB or > logging some action) >
Yes, obviously you wouldn't want to rely on only the HTTP method if you're writing to the same database within GET requests. I would suggest if you are already to the point of scaling out the main database with read replicas in this fashion, then maybe sessions and logging messages should go in a separate database altogether (in particular, sessions would be better in Redis). > 2) all web2py requests are wrapped in a transaction, can this cause > deadlocks? > Are you referring to the last proposed solution involving swapping out the DAL adapter? If so, how would that result in a deadlock? Note, there was a problem with the original code -- both DAL instances ended up with the same adapter, so web2py would not close/return the connection of the read-only database. I changed it so the adapters are swapped between the DAL instances, so both connections will be committed/closed/returned at the end of the request. Anyway, I don't think that is the best option, as it is a hack that relies on details of the internal implementation (and may have unforeseen consequences). I was just brainstorming a way to decouple model definitions from database connections. > 3) there is a delay in syncing a master with a slave db, can this result > in inconsistencies on the user side? > Sure, that's always the trade-off with read replicas. It is suggested in the book <http://web2py.com/books/default/chapter/29/13/deployment-recipes#Using-replicated-databases>, though. > > Why do this instead of simply caching read only queries using redis? > It probably depends on the use case. If you have lots of unique user-specific queries and/or query results that change frequently, caching may not relieve much of the burden on the database. I suppose in many cases where the freshness of a read replica is an issue, the freshness of a cache would also likely be an issue. Anthony -- Resources: - http://web2py.com - http://web2py.com/book (Documentation) - http://github.com/web2py/web2py (Source code) - https://code.google.com/p/web2py/issues/list (Report Issues) --- You received this message because you are subscribed to the Google Groups "web2py-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
