This is running in multiple Sidekiq instances and workers at the same time and it seems that is has generated a couple of issues, like instances getting assigned the "It was alerted recently" error when shouldn't and the opposite.
It is rare, but it is happening, is this the problem or maybe it is something else?
class BrokenModel < ActiveRecord::Base
validates_with BrokenValidator
end
class BrokenValidator < ActiveModel::Validator
def validate record
#record = record
check_alerted
end
private
def check_alerted
if AtomicGlobalAlerted.new(#record).valid?
#record.errors[:base] << "It was alerted recently"
end
p "check_alerted: #{#record.errors[:base]}"
end
end
class AtomicGlobalAlerted
include Redis::Objects
attr_accessor :id
def initialize id
#id = id
#fredis = nil
Sidekiq.redis do |redis|
#fredis = FreshRedis.new(redis, freshness: 7.days, granularity: 4.hours)
end
end
def valid?
#fredis.smembers.includes?(#id)
end
end
We were experiencing something similar at work and after A LOT of digging finally figured out what was happening.
The class method validates_with uses one instance of the validator (BrokenValidator) to validate all instances of the class you're trying to validate (BrokenModel). Normally this is fine but you are assigning a variable (#record) and accessing that variable in another method (check_alerted) so other threads are assigning #record while other threads are still trying to check_alerted.
There are two ways you can fix this:
1) Pass record to check_alerted:
class BrokenValidator < ActiveModel::Validator
def validate(record)
check_alerted(record)
end
private
def check_alerted(record)
if AtomicGlobalAlerted.new(record).valid?
record.errors[:base] << "It was alerted recently"
end
p "check_alerted: #{record.errors[:base]}"
end
end
2) Use the instance version of validates_with which makes a new validator instance for each model instance you want to validate:
class BrokenModel < ActiveRecord::Base
validate :instance_validators
def instance_validators
validates_with BrokenValidator
end
end
Either solution should work and solve the concurrency problem. Let me know if you experience any other issues.
I believe there are some thread safety issues in rails but we can overcome them by taking necessary precautions.
The local variables, such as your local var, are local to each particular invocation of the method block. If two threads are calling this block at the same time, then each call will get its own local context variable and those won't overlap unless there are shared resources involved: instance variables like (#global_var), static variables (##static_var), globals ($global_var) can cause concurrency problems.
You are using instance variable, just instantiate it every time you are coming to the validate_record method and hopefully your problem will go away like :
def validate record
#record.errors[:base] = []
#record = record
check_alerted
end
For more details you can visit this detailed link
Or try to study about rails configs here : link
Related
I have the following model:
class TwitterEngagement < ApplicationRecord
end
And I would like to override create (and create!), update (and
update!) methods of it so no one can manually entry fake data. I would like the help of someone more experienced with active record and rails so I don't mess anything up. Right now what I have is:
class TwitterEngagement < ApplicationRecord
belongs_to :page
def create
super(metrics)
end
def update
super(metrics)
end
private
def metrics
client.get_engagements(page.url)
def client
TwitterClient.new
end
end
Thank you.
TL;DR:
class FacebookEngagement < ApplicationRecord
def create_or_update(*args, &block)
super(metrics)
end
Probably depends on your Rails version, but I traced the ActiveRecord::Persistence sometime before in Rails 5, and found out that both create and update eventually calls create_or_update.
Suggestion:
If ever possible, I'll just do a validation, because it kinda makes more sense because you are validating the inputs, and then probably set an optional readonly?, to prevent saving of records. This will also prevent "silent failing" code / behaviour as doing TL;DR above would not throw an exception / populate the validation errors, if say an unsuspecting developer does: facebook_engagement.update(someattr: 'somevalue') as the arguments are gonna basically be ignored because it's instead calling super(metrics), and would then break the principle of least surprise.
So, I'll probably do something like below:
class FacebookEngagement < ApplicationRecord
belongs_to :page
validate :attributes_should_not_be_set_manually
before_save :set_attributes_from_facebook_engagement
# optional
def readonly?
# allows `create`, prevents `update`
persisted?
end
private
def attributes_should_not_be_set_manually
changes.keys.except('page_id').each do |attribute|
errors.add(attribute, 'should not be set manually!')
end
end
def set_attributes_from_facebook_engagement
assign_attributes(metrics)
end
def metrics
# simple memoization to prevent wasteful duplicate requests (or remove if not needed)
#metrics ||= graph.get_object("#{page.url}?fields=engagement")
end
def graph
Koala::Facebook::API.new
end
end
I'm not 100% sure about why ActiveModel::Dirty has its name. I'm guessing it is because it is considered as dirty to use it.
But in some cases, it is not possible to avoid watching on specific fields.
Ex:
if self.name_changed?
self.slug = self.name.parameterize
end
Without ActiveModel::Dirty, the code would look like:
if old_name != self.name
self.slug = self.name.parameterize
end
which implies having stored old_name before, and it is not readable, so IMHO, it is dirtier than using ActiveModel::Dirty. It becomes even worse if old_number is a number and equals params[:user]['old_number'] as it needs to be correctly formated (parsed as int), whereas ActiveRecord does this automagically.
So I would find clean to define watchable fields at Model level:
class User < ActiveRecord::Base
include ActiveModel::Dirty
watchable_fields :name
before_save :generate_slug, if: name_changed?
def generate_slug
self.slug = self.name.parameterize
end
end
Or (even better?) at controller level, before assigning new values:
def update
#user = current_user
#user.watch_fields(:name)
#user.assign_attributes(params[:user])
#user.generate_slug if #user.name_changed?
#user.save # etc.
end
The good thing here is that it removes the memory overload produced by using ActiveModel::Dirty.
So my question is:
Can I do that using ActiveRecord pre-built tools, or should I write a custom library to this?
Thanks
If ActiveModel::Dirty solves your problem, feel free to use it. The name comes from the term "dirty objects" and is not meant to imply that it's a dirty/hackish module.
See this answer for more details on dirty objects: What is meant by the term "dirty object"?
Here's what I have ended up doing. I like it:
class User < ActiveRecord::Base
attr_accessor :watched
def field_watch(field_name)
self.watched ||= {}
self.watched[field_name] = self.send(field_name)
return self
end
def field_changed?(field_name)
self.send(field_name) != self.watched(field_name)
end
end
And in the controller
def update
#user = current_user.field_watch(:name)
#user.assign_attributes(params[:user])
#user.generate_slug if #user.field_changed?(:name)
#user.save
end
I'll report here if I take the time to wrap this code in a gem or something.
I want to initialize some attributes in retrieved objects with values received from an external API. after_find and after_initialize callbacks won't work for me as this way I have to call the API for each received object, which is is quite slow. I want something like the following:
class Server < ActiveRecord::Base
attr_accessor :dns_names
...
after_find_collection do |servers|
all_dns_names = ForeignLibrary.get_all_dns_entries
servers.each do |s|
s.dns_names = all_dns_names.select{|r| r.ip == s.ip}.map{|r| r.fqdn}
end
end
end
Please note that caching is not a solution, as I need to always have current data, and the data may be changed outside the application.
You'd want to have a class method that enhances each server found with your data. so, something like:
def index
servers = Server.where(condition: params[:condition]).where(second: params[:second])
#servers = Server.with_domains_names(servers)
end
class Server
def self.with_domain_names(servers)
all_dns_names = ForeignLibrary.get_all_dns_entries
servers.each do |s|
s.dns_names = all_dns_names.select{|r| r.ip == s.ip}.map{|r| r.fqdn}
end
end
end
This way, the ForeignLibrary.get_all_dns_entries only gets run once, and you can enhance your servers with that extra information.
If you wanted to do this every time you initialize a server object, I'd simply delegate rather than use after_initialize. So you'd effectively store the all dns entries in a global variable, and then cache it for a period of time. ForeignLibrary.get_all_dns_entries call. So, it would be something like:
class Server
def dns_names
ForeignLibrary.dns_for_server(self)
end
end
class ForeignLibrary
def self.reset
##all_dns_names = nil
end
def self.dns_for_server(server)
all_dns_names.select{|r| r.ip == server.ip}.map{|r| r.fqdn}
end
def self.all_dns_names
Mutex.new.synchronize do
##all_dns_names ||= call_the_library_expensively
end
end
end
(I also used a mutex here since we are doing ||= with class variables)
to use it, you would:
class ApplicationController
before_filter do
ForeignLibrary.reset #ensure every page load has the absolute latest data
end
end
EDITED:
I trying to refactore this:
self.before_create do |obj|
obj.position = self.class.count
end
to this:
self.before_create :set_position
private
def set_position
obj.position = self.class.count
end
But, Its displaying Error:
undefined local variable or method `obj'
How fix this?
class YourClass < AR
before_create :set_position
private
def set_position
self.position = ##count
end
end
The set_position method is instance method, so self is assigned to the instance. The count I believe should be a class variable.
class Something < ActiveRecord::Base
before_create :set_position
protected
def set_position
self.position = self.class.count
end
end
Be aware that if you use class variables you might run into trouble if your server uses multiple processes (##count getting out of sync between them). A functional style queries count from the db.
Note: A lot of this behavior (including properly implemented atomic changes to lists) can be found in the acts_as_list gem, that used to be part of Rails a while back. I suggest using that.
For example, if I have a user model and I need to validate login only (which can happen when validating a form via ajax), it would be great if I use the same model validations defined in the User model without actually instantiating a User instance.
So in the controller I'd be able to write code like
User.valid_attribute?(:login, "login value")
Is there anyway I can do this?
Since validations operate on instances (and they use the errors attribute of an instance as a container for error messages), you can't use them without having the object instantiated. Having said that, you can hide this needed behaviour into a class method:
class User < ActiveRecord::Base
def self.valid_attribute?(attr, value)
mock = self.new(attr => value)
unless mock.valid?
return mock.errors.has_key?(attr)
end
true
end
end
Now, you can call
User.valid_attribute?(:login, "login value")
just as you intended.
(Ideally, you'd include that class method directly into the ActiveRecord::Base so it would be available to every model.)
Thank you Milan for your suggestion. Inspired by it I created a simple module one can use to add this functionality to any class. Note that the original Milans suggestion has a logic error as line:
return mock.errors.has_key?(attr)
should clearly be:
return (not mock.errors.has_key?(attr))
I've tested my solution and it should work, but ofc I give no guarantees. And here's my glorious solution. Basically a 2-liner if you take away the module stuff.. It accepts method names as stings or symbols.
module SingleAttributeValidation
def self.included(klass)
klass.extend(ClassMethods)
end
module ClassMethods
def valid_attribute?(attr, value)
mock = self.new(attr => value)
(not mock.valid?) && (not mock.errors.has_key?(attr.class == Symbol ? attr : attr.to_sym))
end
end
end
To use your standard validation routines:
User.new(:login => 'login_value').valid?
If that does not work for you, build a custom class method for this:
class User < ActiveRecord::Base
validate do |user|
user.errors.add('existing') unless User.valid_login?(user.login)
end
def self.valid_login?(login)
# your validation here
!User.exist?(:login=> login)
end
end
I had a hell of a time getting this to work in Rails 3.1. This finally worked. (Not sure if it's the best way to do it, I'm kind of a newb.). The problem I was having was that value was being set to type ActiveSupport::SafeBuffer, and was failing validation.
def self.valid_attribute?(attr, value)
mock = User.new(attr => "#{value}") # Rails3 SafeBuffer messes up validation
unless mock.valid?
return (not mock.errors.messages.has_key?(attr))
end
return true
end
I have gone with the custom class solution but I just wanted to make sure there was no better way
class ModelValidator
def self.validate_atrribute(klass, attribute, value)
obj = Klass.new
obj.send("#{attribute}=", value)
obj.valid?
errors = obj.errors.on(attribute).to_a
return (errors.length > 0), errors
end
end
and I can use it like
valid, errors = ModelValidator.validate_attribute(User, "login", "humanzz")
class User < ActiveRecord::Base
validates_each :login do |record, attr, value|
record.errors.add attr, 'error message here' unless User.valid_login?(value)
end
def self.valid_login?(login)
# do validation
end
end
Just call User.valid_login?(login) to see if login itself is valid
An implementation of the 'valid_attribute' method you are suggesting:
class ActiveRecord:Base
def self.valid_attribute?(attribute, value)
instance = new
instance[attribute] = value
instance.valid?
list_of_errors = instance.errors.instance_variable_get('#errors')[attribute]
list_of_errors && list_of_errors.size == 0
end
end
How about:
User.columns_hash.has_key?('login')