Failure Self Defense 1 Defend your App against failures in a (micro) services world
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Failure Self Defense
1
Defend your App against failures in a (micro) services world
Tony Fabeen
APP
APP
CRM
APP
CRM
APP
CRM
BillingPayments
APP
CRM
BillingPayments
Database
Search
Queue
APP
CRM
BillingPayments
Database
Search
Queue
WELCOME TO THE
DEV PARTY
Database
Database
APPAPPAPPAPP
APPAPPAPPAPP
APP
Products
Orders
Payments
module Api
class Products
def initialize(host = nil)
@client =
HTTPClient.new(host)
end
def all
@client.get('/products')
end
end
end
def send_checkout
params = { email: @email, token: @token,
ssl_version: :SSLv3 }
RestClient.post(checkout_url, checkout_xml,
params: params,
content_type: "application/xml"){|resp, request, result|
resp }
end
APP Orders
Payments
Orders
Payments
CASCADING FAILURE
MAP DEPENDENCIES
THAT IMPACTS YOUR SYSTEM
Products
Orders
Payments
Catalog
Checkout
Pack
Route
Authorize
Charge
Feature Service
Products
Orders
Payments
Catalog
Checkout
Pack
Route
Authorize
Charge
Degraded
Down
Up
Feature Service
Products
Orders
Payments
Catalog
Checkout
Pack
Route
Authorize
Charge
Degraded
Down
Up
Feature Service
Products
Orders
Payments
Catalog
Checkout
Pack
Route
Authorize
Charge
Degraded
Down
Up
Feature Service
Toxiproxy.populate([{ name: 'redis', listen: '127.0.0.1:22222', upstream: '127.0.0.1:6379' }])
context 'when service UP' do
before { Cache.put('key', 'value') }
it 'saves value' do
expect(Cache.get('key')).to eq('value')
end
end
context 'when service DOWN' do
it 'will raises error' do
Toxiproxy[:redis].down do
expect { Cache.put('key', 'value') }.to
raise_error(Redis::CannotConnectError)
end
end
end
context 'when service UP' do
before { Cache.put('key', 'value') }
it 'saves value' do
expect(Cache.get('key')).to eq('value')
end
end
context 'when service DOWN' do
it 'will raises error' do
Toxiproxy[:redis].down do
expect { Cache.put('key', 'value') }.to
raise_error(Redis::CannotConnectError)
end
end
end
Fault Tolerance
An application with an average Response Time of 60ms can process 1.000 Requests Per Minute (RPM) per Thread.
An application with an average Response Time of 60ms can process 1.000 Requests Per Minute (RPM) per Thread.
How many Threads we need to handle 100.000 RPM of Throughput ?
100
Imagine that 1% of the traffic timeout on a Service after 30 seconds, the Response Time will raise to 360 ms.
Imagine that 1% of the traffic timeout on a Service after 30 seconds, the Response Time will raise to 360 ms.
How many Threads we need to handle 100.000 RPM of Throughput ?
600
Service
600 RPM0.01 s
0.01 s
0.01 s
0.01 s
0.01 s
0.01 s
Service
60 RPM0.01 s
0.01 s
0.10 s
0.10 s
0.10 s
0.10 s
Service
60 RPM0.01 s
0.01 s
0.10 s
0.10 s
0.10 s
0.10 s
High Response Time
Less Throughput
Fail Fast
Fail Fast
Low timeouts
Fail Fast
Low timeouts
Connection timeout
Fail Fast
Low timeouts
Connection timeout Socket Read timeout
Fail Fast
Low timeouts
Connection timeout Socket Read timeout Resource aquisition
Fail Gracefully
class Cache
def self.put(key, value)
service.set(key, value)
end
def self.get(key)
service.get(key)
end
end
end
class Cache
def self.put(key, value)
service.set(key, value)
end
def self.get(key)
service.get(key)
end
end
end
Cache.put('key', 'value')
Cache.get('key
')
def put(key, value)
service.set(key, value)
true
rescue Redis::CannotConnectError => error
AwesomeLogger.log(error)
false
end
def get(key, fallback_value = nil)
service.get(key)
rescue Redis::CannotConnectError => error
AwesomeLogger.log(error)
fallback_value
end
end
Don't try if you can't succeed
Circuit Breakers
Client Circuit Breaker Service
Closed
Client Circuit Breaker
Closed
Client Circuit Breaker
Open
Error
Client Circuit Breaker
Open
Client Circuit Breaker
Open
Error
Client Circuit Breaker Service
Closed
Slow services
High timeouts
Bulkheads
FeatureA
Bulkhead
2 available
FeatureC
FeatureB
Bulkhead
FeatureC
1 available
FeatureA
FeatureB
FeatureC
no requests available
FeatureA
FeatureB
Bulkhead
Bulkhead
FeatureC
no requests available
FeatureA
FeatureB
Error
Monitor Service Calls
Monitor Service Calls
Timeout rate Rejected call rate
Short circuit rate Failure/Success rate
Response Times
SummaryKnow your dependenciesImprove your test suite
Fail FastTimeouts
Fail GracefullyFallbacks
Don't try if you can't succeedCircuit Breakers and Bulkheads are friends
Monitor Service Calls
Notice problems
?
https://github.com/tonyfabeen
https://twitter.com/tonyfabeen
https://linkedin.com/tonyfabeen
Related Documents
