Hacker News new | past | comments | ask | show | jobs | submit login

Great post! I had a few questions:

1. Do the slaves write to disk? I assume that would incur no performance penalty, as I believe slave writes don't slow down the master's performance -- only the rate of replication.

2. Async operations seem like a great idea, and in fact I was toying around with implementing something like that myself. How far along is your implementation, and how does it seem to behave so far?

3. About your proposed 'automatic slave read' idea: The way I understand what you wrote, it seems like you would end up incurring the same cost on the master and the slave, i.e., you make the request on the server, and if it takes longer than X msecs, you then issue the request on the slave. Instead, is there any merit to having a redis command that returns how much time is likely going to be required for a given command? It wouldn't have to be very precise -- I imagine even the rough order of magnitude would be helpful for deciding whether to get the data from a given master or its slave?




#1: currently no, but I'm about to change that and have them do a SAVE every few hours or so.

#2: my implementation is pretty far along in my testing but I do need to polish it up, update the docs, and test some edge cases

#3: the idea with the slave reads for stuff that is CPU expensive or could slow down other requests (doing large multi-set operations, for example)




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: