In edge rails, serialized attributes are saved every time no matter
they are changed or not:
def update_with_dirty
if partial_updates?
# Serialized attributes should always be written in case
they’ve been
# changed in place.
update_without_dirty(changed | (attributes.keys &
self.class.serialized_attributes.keys))
else
update_without_dirty
end
end
In out app, User model has a serialized attribute friend_ids, which is
the id array of the user’s friends (used to be a friendships table,
but when that table grew to tens of millions of records, we refactored
it to a User’s attribute). The User model in our app is saved very
frenquently, and many users have more than 1 handred friends, so the
friend_ids may has 1kB long. Some of my colleages opposed to use
serialized friend_ids, and suggested to used a comma separated string
instead because saving 1kB each time would have performance issue.
instead because saving 1kB each time would have performance issue.
Well before rails 2.1 partial updates didn’t exist at all and the
world continued to function
The simple to solution is to benchmark it - monkey patch stuff so that
this does happen, measure, monkey patch it so that is doesn’t, measure
again.
Another possible way would be to stick the friend_ids data in its own
table. This might also save you loading it unecessarily
The world continued to function because few people used rails in high
profile website at that time. Our app has more than 50 million HTTP
requests per day, so we have to hack a lot of rails plugins that have
performance issues under such load:(
The patch proposed here would fix this issue: https://rails.lighthouseapp.com/projects/8994/tickets/2764-supporting-partial-updates-for-serialized-columns
I hope it will be merged to edge rails
The world continued to function because few people used rails in high
profile website at that time. Our app has more than 50 million HTTP
requests per day, so we have to hack a lot of rails plugins that have
performance issues under such load:(
The patch proposed here would fix this issue:#2764 Supporting partial updates for serialized columns - Ruby on Rails - rails…
I hope it will be merged to edge rails
No, actually, it won’t. Unless you’re claiming that the bottleneck is
in transmitting the data to the DB, that patch still converts every
serialized attribute to YAML almost every time that the record is
saved (note the calls to object_to_yaml). Serialized fields that are
changed will get converted again. So if anything, the overhead may
be worse than before…
I also find the idea of taking one of the primary parts of a social
networking site (friend linking) and making it opaque through
serialization to be very odd. Isn’t the whole point of the “social
media” fad to be analyzing and drawing conclusions from the network
graph? Meh.
Transforming to/from yml consumes CPU cycles, but storing a large
field to database consumes network bandwidth and disk IO:(
In this case, I am afraid my colleague’s solution is right: save that
list as a comma separate string.
Your 2nd point is right, I will consider not storing friend list
later. But besides friend list, there are other fields that need to
stored as serialized attr.
Thanks
This forum is not affiliated to the Ruby language, Ruby on Rails framework, nor any Ruby applications discussed here.
