• A common language (for the two ends to understand)
• A way to address who do you want to communicate with
• A Connection (so the content of of the communication can reach the recipients)

A set of protocols that define how two or more devices can communicate with each other. To learn more about TCP/IP, read here

Ethernet simply refers to the most common type of Local Area Network (LAN) used today. A LAN—in contrast to a WAN (Wide Area Network), which spans a larger geographical area—is a connected network of computers in a small area, like your office, college campus, or even home.

When a device sends a packet to the broadcast MAC address (FF:FF:FF:FF:FF:FF​), it is delivered to all stations on the local network. Ethernet broadcasts are used to resolve IP addresses to MAC addresses (by ARP) at the datalink layer .

An Internet Protocol address (IP address) is a numerical label assigned to each device connected to a computer network that uses the Internet Protocol for communication.An IP address serves two main functions: host or network interface identification and location addressing.

You can read more about the OSI model in penguintutor.com

Multicast: Sending a message to a group of subscribers. It can be one-to-many or many-to-many.

1. Before sending a frame, it checks whether another host already transmitting a frame.
2. If no one transmitting, it starts transmitting the frame.
3. If two hosts transmitted at the same time, we have a collision.
4. Both hosts stop sending the frame and they send to everyone a 'jam signal' notifying everyone that a collision occurred
5. They are waiting for a random time before sending again
6. Once each host waited for a random time, they try to send the frame again and so the

Proxy servers provide varying levels of functionality, security, and privacy depending on your use case, needs, or company policy.

• A client node sends a SYN data packet over an IP network to a server on the same or an external network. The objective of this packet is to ask/infer if the server is open for new connections.
• The target server must have open ports that can accept and initiate new connections. When the server receives the SYN packet from the client node, it responds and returns a confirmation receipt – the ACK packet or SYN/ACK packet.
• The client node receives the SYN/ACK from the server and responds with an ACK packet.

Bonus question: what is the RTT of LAN?

Penguintutor.com provides a good explanation.

A default gateway serves as an access point or IP router that a networked computer uses to send information to a computer in another network or the internet.

Systems keep an ARP look-up table where they store information about what IP addresses are associated with what MAC addresses. When trying to send a packet to an IP address, the system will first consult this table to see if it already knows the MAC address. If there is a value cached, ARP is not used.

Read more here

NAT stands for network address translation. It’s a way to map multiple local private addresses to a public one before transferring the information. Organizations that want multiple devices to employ a single IP address use NAT, as do most home routers. For example, your computer's private IP could be 192.168.1.100, but your router maps the traffic to it's public IP (e.g. 1.1.1.1). Any device on the internet would see the traffic coming from your public IP (1.1.1.1) instead of your private IP (192.168.1.100).

• SSH - 22
• SMTP - 25
• HTTP - 80
• DNS - 53
• HTTPS - 443
• FTP - 21
• SFTP - 22

The control plane is the part of the network that decides how to route and forward packets to a different location.

The data plane is the part of the network that actually forwards the data/packets.

Refers to monitoring and management functions.

Control Plane.

Latency is the time taken for an information to reach its destination from the source.

Bandwidth is the capacity of a communication channel to measure how much data the latter can handle over a specific time period. More bandwidth would imply more traffic handling and thus more data transfer.

Throughput refers to the measurement of the real amount of data transferred over a certain period of time across any transmission channel.

Latency. To have a good latency, a search query should be forwarded to the closest datacenter.

Throughput. To have a good throughput, the upload stream should be routed to an underutilized link.

• Keep caches updated (which means the request could be forwarded not to the closest datacenter)

00110011110100011101

• Hypertext Transfer Protocol (HTTP) - used for the webpages on the internet
• Simple Mail Transfer Protocol (SMTP) - email transmission
• Telecommunications Network - (TELNET) - terminal emulation to allow client access to telnet server
• File Transfer Protocol (FTP) - facilitates transfer of files between any two machines
• Domain Name System (DNS) - domain name translation
• Dynamic Host Configuration Protocol (DHCP) - allocates IP addresses, subnet masks and gateways to hosts
• Simple Network Management Protocol (SNMP) - gathers data of devices on the network

• Internet Protocol (IP) - assists in routing packets from one machine to another
• Internet Control Message Protocol (ICMP) - lets one know what is going such as error messages and debugging information

Read more [here](https://www.globalsign.com/en/blog/what-is-hsts-and-how-do-i-use-it#:~:text=HTTP%20Strict%20Transport%20Security%20(HSTS,and%20back%20to%20the%20browser.)

The internet refers to network of networks, transferring huge amounts of data around the globe.
The World Wide Web is an application running on millions of server, on top of the internet, accessed through what is know as the web browser

ISP (Internet Service Provider) is the local internet company provider.

"responsible for making it easy to run programs (even allowing you to seemingly run many at the same time), allowing programs to share memory, enabling programs to interact with devices, and other fun stuff like that".

A process is a running program. A program is one or more instructions and the program (or process) is executed by the operating system.

• Create - allow to create new processes
• Delete - allow to remove/destroy processes
• State - allow to check the state of the process, whether it's running, stopped, waiting, etc.
• Stop - allow to stop a running process

• The OS is reading program's code and any additional relevant data
• Program's code is loaded into the memory or more specifically, into the address space of the process.
• Memory is allocated for program's stack (aka run-time stack). The stack also initialized by the OS with data like argv, argc and parameters to main()
• Memory is allocated for program's heap which is required for dynamically allocated data like the data structures linked lists and hash tables
• I/O initialization tasks are performed, like in Unix/Linux based systems where each process has 3 file descriptors (input, output and error)
• OS is running the program, starting from main()

False. It was true in the past but today's operating systems perform lazy loading which means only the relevant pieces required for the process to run are loaded first.

• Running - it's executing instructions
• Ready - it's ready to run but for different reasons it's on hold
• Blocked - it's waiting for some operation to complete. For example I/O disk request

• I/O operations (e.g. Reading from a disk)
• Waiting for a packet from a network

Even when using a system with one physical CPU, it's possible to allow multiple users to work on it and run programs. This is possible with time sharing where computing resources are shared in a way it seems to the user the system has multiple CPUs but in fact it's simply one CPU shared by applying multiprogramming and multi-tasking.

Somewhat the opposite of time sharing. While in time sharing a resource is used for a while by one entity and then the same resource can be used by another resource, in space sharing the space is shared by multiple entities but in a way where it's not being transferred between them.
It's used by one entity until this entity decides to get rid of it. Take for example storage. In storage, a file is yours until you decide to delete it.

CPU scheduler

• Allocating memory
• Schedule processes
• Control CPU

True

Buffer: Reserved place in RAM which is used to hold data for temporary purposes Cache: Cache is usually used when processes reading and writing to the disk to make the process faster by making similar data used by different programs easily accessible.

Virtualization uses software to create an abstraction layer over computer hardware that allows the hardware elements of a single computer—processors, memory, storage and more - to be divided into multiple virtual computers, commonly called virtual machines (VMs).

Read more here

Hosted hypervisors and bare-metal hypervisors.

On the other hand, there will probably be some limitation regarding loading (any) drivers so a hosted hypervisor will usually benefit from having a better hardware compatibility.

Operating system virtualization Network functions virtualization Desktop virtualization

Yes, it's a operating-system-level virtualization, where the kernel is shared and allows to use multiple isolated user-spaces instances.

The introduction of virtual machines allowed companies to deploy multiple business applications on the same hardware while each application is separated from each other in secured way, where each is running on its own separate operating system.

1. Translation lookup in i18n
2. Hold the result of the last executed expression or statement in the interactive interpreter.
3. As a general purpose "throwaway" variable name. For example: x, y, _ = get_data() (x and y are used but since we don't care about third variable, we "threw it away").

A tuple is a built-in data type in Python. It's used for storing multiple items in a single variable.

List, as opposed to a tuple, is a mutable data type. It means we can modify it and at items to it.

x.append(2)

some_list[-1]

Don't use append unless you would like the list as one item.

my_list[0:3] = []

• sorted(list) will return a new list (original list doesn't change)

• list.sort() will return None but the list is change in-place

• sorted() works on any iterable (Dictionaries, Strings, ...)

• list.sort() is faster than sorted(list) in case of Lists

There are many ways of solving this problem:
# Note: :list and -> bool are just python typings, they are not needed for the correct execution of the algorithm.

Taking advantage of sets and len:

def is_unique(l:list) -> bool:
    return len(set(l)) == len(l)

This one is can be seen used in other programming languages.

def is_unique2(l:list) -> bool:
    seen = []

    for i in l:
        if i in seen:
            return False
        seen.append(i)
    return True

Here we just count and make sure every element is just repeated once.

def is_unique3(l:list) -> bool:
    for i in l:
        if l.count(i) > 1:
            return False
    return True

This one might look more convulated but hey, one liners.

def is_unique4(l:list) -> bool:
    return all(map(lambda x: l.count(x) < 2, l))

my_dict = dict(x=1, y=2) OR my_dict = {'x': 1, 'y': 2} OR my_dict = dict([('x', 1), ('y', 2)])

del my_dict['some_key'] you can also use my_dict.pop('some_key') which returns the value of the key.

Solution

Using the re module

It will reverse the string, so x would be equal to azzip.

mario

for i in range(3, 3):
   print(i)

No output :)

the_list.sort(key=lambda x: x[1])

pdb :D

Short answer is: It returns a None object.

We could go a bit deeper and explain the difference between

def a ():
    return

>>> None

And

def a ():
    pass

>>> None

Or we could be asked this as a following question, since they both give the same result.

We could use the dis module to see what's going on:

  2           0 LOAD_CONST               0 (<code object a at 0x0000029C4D3C2DB0, file "<dis>", line 2>)
              2 LOAD_CONST               1 ('a')
              4 MAKE_FUNCTION            0
              6 STORE_NAME               0 (a)

  5           8 LOAD_CONST               2 (<code object b at 0x0000029C4D3C2ED0, file "<dis>", line 5>)
             10 LOAD_CONST               3 ('b')
             12 MAKE_FUNCTION            0
             14 STORE_NAME               1 (b)
             16 LOAD_CONST               4 (None)
             18 RETURN_VALUE

Disassembly of <code object a at 0x0000029C4D3C2DB0, file "<dis>", line 2>:
  3           0 LOAD_CONST               0 (None)
              2 RETURN_VALUE

Disassembly of <code object b at 0x0000029C4D3C2ED0, file "<dis>", line 5>:
  6           0 LOAD_CONST               0 (None)
              2 RETURN_VALUE

An empty return is exactly the same as return None and functions without any explicit return will always return None regardless of the operations, therefore

def sum(a, b):
    global c
    c = a + b

>>> None

def print_list(self): node = self.head while(node): print(node.data) node = node.next

Read about descriptors here

This is a good reference https://docs.python.org/3/library/datatypes.html

def wee(word):
    return word

def oh(f):
    return f + "Ohh"

>>> oh(wee("Wee"))
<<< Wee Ohh

every time, that's what decorators do, they introduce syntax to write all of this on the go, using the keyword '@'.

Google: "Monitoring is one of the primary means by which service owners keep track of a system’s health and availability".

This approach require from a human to always check why the value exceeded and how to handle it while today, it is more effective to notify people only when they need to take an actual action. If the issue doesn't require any human intervention, then the problem can be fixed by some processes running in the relevant environment.

Alerts
Tickets
Logging

From Prometheus documentation: "if you need 100% accuracy, such as for per-request billing".

Prometheus server is responsible for scraping and storing the data
Push gateway is used for short-lived jobs
Alert manager is responsible for alerts ;)

Go also has good community.

With var x int = 2 we are setting the variable type to integer while with x := 2 we are letting Go figure out by itself the type.

False. We can't redeclare variables but yes, we must used declared variables.

• fmt - formatted I/O

package main

import "fmt"

func main() {
    var x int = 101
    var y string
    y = string(x)
    fmt.Println(y)
}

It looks what unicode value is set at 101 and uses it for converting the integer to a string. If you want to get "101" you should use the package "strconv" and replace y = string(x) with y = strconv.Itoa(x)

package main

func main() {
    var x = 2
    var y = 3
    const someConst = x + y
}

Constants in Go can only be declared using constant expressions. But x, y and their sum is variable.
const initializer x + y is not a constant

package main

import "fmt"

const (
	x = iota
	y = iota
)
const z = iota

func main() {
	fmt.Printf("%v\n", x)
	fmt.Printf("%v\n", y)
	fmt.Printf("%v\n", z)
}

Go's iota identifier is used in const declarations to simplify definitions of incrementing numbers. Because it can be used in expressions, it provides a generality beyond that of simple enumerations.
x and y in the first iota group, z in the second.
Iota page in Go Wiki

It avoids having to declare all the variables for the returns values. It is called the blank identifier.
answer in SO

package main

import "fmt"

const (
	_ = iota + 3
	x
)

func main() {
	fmt.Printf("%v\n", x)
}

Since the first iota is declared with the value 3 ( + 3), the next one has the value 4

package main

import (
	"fmt"
	"sync"
	"time"
)

func main() {
	var wg sync.WaitGroup

	wg.Add(1)
	go func() {
		time.Sleep(time.Second * 2)
		fmt.Println("1")
		wg.Done()
	}()

	go func() {
		fmt.Println("2")
	}()

	wg.Wait()
	fmt.Println("3")
}

Golang package sync

package main

import (
	"fmt"
)

func mod1(a []int) {
	for i := range a {
		a[i] = 5
	}

	fmt.Println("1:", a)
}

func mod2(a []int) {
	a = append(a, 125) // !

	for i := range a {
		a[i] = 5
	}

	fmt.Println("2:", a)
}

func main() {
	s1 := []int{1, 2, 3, 4}
	mod1(s1)
	fmt.Println("1:", s1)

	s2 := []int{1, 2, 3, 4}
	mod2(s2)
	fmt.Println("2:", s2)
}

Aritcle about arrays, Blog post about append

package main

import (
	"container/heap"
	"fmt"
)

// An IntHeap is a min-heap of ints.
type IntHeap []int

func (h IntHeap) Len() int           { return len(h) }
func (h IntHeap) Less(i, j int) bool { return h[i] < h[j] }
func (h IntHeap) Swap(i, j int)      { h[i], h[j] = h[j], h[i] }

func (h *IntHeap) Push(x interface{}) {
	// Push and Pop use pointer receivers because they modify the slice's length,
	// not just its contents.
	*h = append(*h, x.(int))
}

func (h *IntHeap) Pop() interface{} {
	old := *h
	n := len(old)
	x := old[n-1]
	*h = old[0 : n-1]
	return x
}

func main() {
	h := &IntHeap{4, 8, 3, 6}
	heap.Init(h)
	heap.Push(h, 7)

  fmt.Println((*h)[0])
}

Golang container/heap package

The main difference is that SQL databases are structured (data is stored in the form of tables with rows and columns - like an excel spreadsheet table) while NoSQL is unstructured, and the data storage can vary depending on how the NoSQL DB is set up, such as key-value pair, document-oriented, etc.

• Heterogeneous data which changes often
• Data consistency and integrity is not top priority
• Best if the database needs to scale rapidly

• mongoexport
• programming languages

SQL (Structured Query Language) is a standard language for relational databases (like MySQL, MariaDB, ...).
It's used for reading, updating, removing and creating data in a relational database.

The main difference is that SQL databases are structured (data is stored in the form of tables with rows and columns - like an excel spreadsheet table) while NoSQL is unstructured, and the data storage can vary depending on how the NoSQL DB is set up, such as key-value pair, document-oriented, etc.

Additionally, since NoSQL does not adhere to the strict table with columns and rows structure that Relational Databases require, you can store different data types together.

Select *
From Customers;

Select Items_in_cart
From Customers
Where Customer_Name = "John Smith";

Select SUM(Cash_spent_to_Date) as SUM_CASH
From Customers;

Select count(1) as Number_of_People_w_items
From Customers
where Items_in_cart > 0;

You would join them on the unique key. In this case, the unique key is Customer_ID in both the Customers table and Orders table

The Sum of all the purchases of cat food weren't readily available, so we used a with statement to create the pseudo table to retrieve the sum of the prices spent by each customer, then join the table normally.

SELECT count(*)                             SELECT count(*)
FROM shawarma_purchases                     FROM shawarma_purchases
WHERE                               vs.     WHERE
  YEAR(purchased_at) == '2017'              purchased_at >= '2017-01-01' AND
                                            purchased_at <= '2017-31-12'

SELECT count(*)
FROM shawarma_purchases
WHERE
  purchased_at >= '2017-01-01' AND
  purchased_at <= '2017-31-12'

When you use a function (YEAR(purchased_at)) it has to scan the whole database as opposed to using indexes and basically the column as it is, in its natural state.

startap-script

• It is the managed kubernetes service on GCP for deploying, managing and scaling containerised applications using Google infrastructure.

fun fact: Anthos is flower in greek, they grow in the ground (earth) but need rain from the clouds to flourish.

• Infrastructure management - Google Kubernetes Engine (GKE)
• Cluster management - GKE, Ingress for Anthos
• Service management - Anthos Service Mesh
• Policy enforcement - Anthos Config Management, Anthos Enterprise Data Protection, Policy Controller
• Application deployment - CI/CD tools like Cloud Build, GitLab
• Application development - Cloud Code

• Google Kubernetes Engine (GKE)

On GCP the kubernetes api-server is the only control plane component exposed to customers whilst compute engine manages instances in the project.

• Networking load balancing for L4 and HTTP(S) Load Balancing for L7 which are both managed services that do not require additional configuration.
• Ingress for Anthos which allows the ability to deploy a load balancer that serves an application across multiple clusters on GKE

• Yes, Anthos on AWS is now GA. For more read here

• Control plane security - GCP manages and maintains the K8s control plane out of the box. The user can secure the api-server by using master authorized networks and private clusters. These allow the user to disable access on the public IP address by assigning a private IP address to the master.
• Node security - By default workloads are provisioned on Compute engine instances that use Google's Container Optimised OS. This operating system implements a locked-down firewall, limited user accounts with root disabled and a read-only filesystem. There is a further option to enable GKE Sandbox for stronger isolation in multi-tenant deployment scenarios.
• Network security - Within a created cluster VPC, Anthos GKE leverages a powerful software-defined network that enables simple Pod-to-Pod communications. Network policies allow locking down ingress and egress connections in a given namespace. Filtering can also be implemented to incoming load-balanced traffic for services that require external access, by supplying whitelisted CIDR IP ranges.
• Workload security - Running workloads run with limited privileges, default Docker AppArmor security policies are applied to all Kubernetes Pods. Workload identity for Anthos GKE aligns with the open source kubernetes service accounts with GCP service account permissions.
• Audit logging - Adminstrators are given a way to retain, query, process and alert on events of the deployed environments.

• Google Cloud Virtual Private Network (Cloud VPN) - this is for secure networking
• Google Cloud Key Management Service (Cloud KMS) - for key management

• This is when pods can directly talk to each other within a cluster, but cannot be reached from outside the cluster thus forming an "island" within the network that is not connected to the external network.

1. An importer that reads from a central Git repository
2. A component that synchronises stored configuration data into K8s objects
3. A component that monitors drift between desired and actual cluster configurations with a capability of reconciliation when need rises.

It follows common modern software development practices which makes cluster configuration, management and policy changes auditable, revertable, and versionable easily enforcing IT governance and unifying resource management in an organisation.

• It is a suite of tools that assist in monitoring and managing deployed services on Anthos of all shapes and sizes whether running in cloud, hybrid or multi-cloud environments. It leverages the APIs and core components from Istio, a highly configurable and open-source service mesh platform.

1. Data plane - it consists of a set of distributed proxies that mediate all inbound and outbound network traffic between individual services which are configured using a centralised control plane and an open API
2. Control plane - is a fully managed offering outside of Anthos GKE clusters to simplify management overhead and ensure highest possible availability.

1. Traffic Director - it is GCP's fully managed service mesh traffic control plane, responsible for translating Istio API objects into configuration information for the distributed proxies, as well as directing service mesh ingress and egress traffic
2. Managed CA - is a centralised certificate authority responsible for providing SSL certificates to each of the distributed proxies, authentication information and distributing secrets
3. Operations tooling - formerly stackdriver, provides a managed ingestion point for observability and telemetry, specifically monitoring, tracing and logging data generated by each of the proxies. This powers the observability dashboard for operators to visually inspect their services and service dependencies assisting in the implementation of SRE best practices for monitoring SLIs and establishing SLOs.

• Uniform observability - the data plane reports service to service communication back to the control plane generating a service dependency graph. Traffic inspection by the proxy inserts headers to facilitate distributed tracing, capturing and reporting service logs together with service-level metrics (i.e latency, errors, availability).
• Operational agility - fine-grained controls for managing the flow of inter-mesh (north-south) and intra-mesh (east-west) traffic are provided.
• Policy-driven security - policies can be enforced consistently across diverse protocols and runtimes as service communications are secured by default.

• Traffic splitting across differing service versions for canary or A/B testing
• Circuit breaking to prevent cascading failures
• Fault injection to help build resilient and fault-tolerant deployments
• HTTP header-based traffic steering between individual services or versions

It is part of the Anthos stack that brings a serverless container experience to Anthos, offering a high-level platform experience on top of K8s clusters. It is built with Knative, an open-source operator for K8s that brings serverless application serving and eventing capabilities.

• Easy migration from K8s deployments - Without Cloud Run, platform engineers have to configure deployment, service, and HorizontalPodAutoscalers(HPA) objects to a loadbalancer and autoscaling. If application is already serving traffic it becomes hard to change configurations or roll back efficiently. Using Cloud Run all this is managed thus the Knative service manifest describes the application to be autoscaled and loadbalanced
• Autoscaling - a sudden traffic spike may cause application containers in K8s to crash due to overload thus an efficient automated autoscaling is executed to serve the high volume of traffic
• Networking - it has built-in load balancing capabilities and policies for traffic splitting between multiple versions of an application.
• Releases and rollouts - supports the notion of the Knatibe API's revisions which describe new versions or different configurations of your application and canary deployments by splitting traffic.
• Monitoring - observing and recording metrics such as latency, error rate and requests per second.

• Rapid, request-based autoscaling - default autoscalers monitor request metrics which allows Cloud Run for Anthos to handle spiky traffic patterns smoothly
• Concurrency controls - limits such as max in-flight requests per container are enforced to ensure the container does not become overloaded and crash. More containers are added to handle the spiky traffic, buffering the requests.
• Scale to zero - if an application is inactive for a while Cloud Run scales it down to zero to reduce its footprint. Alternatively one can turn off scale-to-zero to prevent cold starts.

• Machine learning model predictions e.g Tensorflow serving containers
• API gateways, API middleware, web front ends and Microservices
• Event handlers, ETL

• Nova - Manage virtual instances
• Cinder - Block Storage
• Keystone - Authentication service across the cloud

• Glance - Images Service. Also used for copying or snapshot instances
• Horizon - GUI for viewing and modifying resources
• Cinder - Block Storage
• Nova - Manage virtual instances

You can read about TripleO right here

• Used to provision and manage virtual instances
• It supports Multi-Tenancy in different levels - logging, end-user control, auditing, etc.
• Highly scalable
• Authentication can be done using internal system or LDAP
• Supports multiple types of block storage
• Tries to be hardware and hypervisor agnostice

• nova-api - the server which serves metadata and compute APIs
• the different Nova components communicate by using a queue (Rabbitmq usually) and a database
• a request for creating an instance is inspected by nova-scheduler which determines where the instance will be created and running
• nova-compute is the component responsible for communicating with the hypervisor for creating the instance and manage its lifecycle

• One of the core component of OpenStack and a standalone project
• Neutron focused on delivering networking as a service
• With Neutron, users can set up networks in the cloud and configure and manage a variety of network services
• Neutron interacts with:
  • Keystone - authorize API calls
  • Nova - nova communicates with neutron to plug NICs into a network
  • Horizon - supports networking entities in the dashboard and also provides topology view which includes networking details

• neutron-l3-agent - L3/NAT forwarding (provides external network access for VMs for example)
• neutron-dhcp-agent - DHCP services
• neutron-metering-agent - L3 traffic metering
• neutron-*-agtent - manages local vSwitch configuration on each compute (based on chosen plugin)
• neutron-server - exposes networking API and passes requests to other plugins if required

• Management Network - used for internal communication between OpenStack components. Any IP address in this network is accessible only within the datacetner
• Guest Network - used for communication between instances/VMs
• API Network - used for services API communication. Any IP address in this network is publicly accessible
• External Network - used for public communication. Any IP address in this network is accessible by anyone on the internet

There are many reasons for that. One for example: you can't remove router if there are active ports assigned to it.

• Glance is the OpenStack image service
• It handles requests related to instances disks and images
• Glance also used for creating snapshots for quick instances backups
• Users can use Glance to create new images or upload existing ones

• glance-api - responsible for handling image API calls such as retrieval and storage. It consists of two APIs: 1. registry-api - responsible for internal requests 2. user API - can be accessed publicly
• glance-registry - responsible for handling image metadata requests (e.g. size, type, etc). This component is private which means it's not available publicly
• metadata definition service - API for custom metadata
• database - for storing images metadata
• image repository - for storing images. This can be a filesystem, swift object storage, HTTP, etc.

• Swift is Object Store service and is an highly available, distributed and consistent store designed for storing a lot of data
• Swift is distributing data across multiple servers while writing it to multiple disks
• One can choose to add additional servers to scale the cluster. All while swift maintaining integrity of the information and data replications.

Not by default. Object Storage API limits the maximum to 5GB per object but it can be adjusted.

• Container - Defines a namespace for objects.
• Account - Defines a namespace for containers
• Object - Data content (e.g. image, document, ...)

False. Two objects can have the same name if they are in different containers.

• Cinder is OpenStack Block Storage service
• It basically provides used with storage resources they can consume with other services such as Nova
• One of the most used implementations of storage supported by Cinder is LVM
• From user perspective this is transparent which means the user doesn't know where, behind the scenes, the storage is located or what type of storage is used

• cinder-api - receives API requests
• cinder-volume - manages attached block devices
• cinder-scheduler - responsible for storing volumes

• Role - A list of rights and privileges determining what a user or a project can perform
• Tenant/Project - Logical representation of a group of resources isolated from other groups of resources. It can be an account, organization, ...
• Service - An endpoint which the user can use for accessing different resources
• Endpoint - a network address which can be used to access a certain OpenStack service
• Token - Used for access resources while describing which resources can be accessed by using a scope

• Name
• ID number
• Type
• Description

• PublicURL - Publicly accessible through public internet
• InternalURL - Used for communication between services
• AdminURL - Used for administrative management

A list of services and their endpoints

• Swift - highly available, distributed, eventually consistent object/blob store
• Sahara - Manage Hadoop Clusters
• Ironic - Bare Metal Provisioning
• Trove - Database as a service that runs on OpenStack
• Aodh - Alarms Service
• Ceilometer - Track and monitor usage

• Database as a service which runs on OpenStack - Trove
• Bare Metal Provisioning - Ironic
• Track and monitor usage - Ceilometer
• Alarms Service - Aodh
• Manage Hadoop Clusters
• Manage Hadoop Clusters - Sahara
• highly available, distributed, eventually consistent object/blob store - Swift

• You can't have OpenStack deployed without Keystone
• It Provides identity, policy and token services
  • The authentication provided is for both users and services
  • The authorization supported is token-based and user-based.
• There is a policy defined based on RBAC stored in a JSON file and each line in that file defines the level of access to apply

• There is a service API and admin API through which Keystone gets requests
• Keystone has four backends:
  • Token Backend - Temporary Tokens for users and services
  • Policy Backend - Rules management and authorization
  • Identity Backend - users and groups (either standalone DB, LDAP, ...)
  • Catalog Backend - Endpoints
• It has pluggable environment where you can integrate with:
  • LDAP
  • KVS (Key Value Store)
  • SQL
  • PAM
  • Memcached

• Keystone gets a call/request and checks whether it's from an authorized user, using username, password and authURL
• Once confirmed, Keystone provides a token.
• A token contains a list of user's projects so there is no to authenticate every time and a token can submitted instead

• nova-api - responsible for managing requests/calls
• nova-compute - responsible for managing instance lifecycle
• nova-conductor - Mediates between nova-compute and the database so nova-compute doesn't access it directly

• Nova-novncproxy - Access through VNC connections
• Nova-spicehtml5proxy - Access through SPICE
• Nova-xvpvncproxy - Access through a VNC connection

• Django-based project focusing on providing an OpenStack dashboard and the ability to create additional customized dashboards
• You can use it to access the different OpenStack services resources - instances, images, networks, ...
  • By accessing the dashboard, users can use it to list, create, remove and modify the different resources
• It's also highly customizable and you can modify or add to it based on your needs

• API is backward compatible
• There are three type of dashboards: user, system and settings
• It provides core support for all OpenStack core projects such as Neutron, Nova, etc. (out of the box, no need to install extra packages or plugins)
• Anyone can extend the dashboards and add new components
• Horizon provides templates and core classes from which one can build its own dashboard

Elasticserach, Logstash and Kibana are also known as the ELK stack.

"Elasticsearch is a distributed document store. Instead of storing information as rows of columnar data, Elasticsearch stores complex data structures that have been serialized as JSON documents"

"Logstash is a powerful, flexible pipeline that collects, enriches and transports data. It works as an extract, transform & load (ETL) tool for collecting log messages."

Beats are lightweight data shippers. These data shippers installed on the client where the data resides. Examples of beats: Filebeat, Metricbeat, Auditbeat. There are much more.

"Kibana is an open source analytics and visualization platform designed to work with Elasticsearch. You use Kibana to search, view, and interact with data stored in Elasticsearch indices. You can easily perform advanced data analysis and visualize your data in a variety of charts, tables, and maps."

1. The data logged by the application is picked by filebeat and sent to logstash
2. Logstash process the log based on the defined filters. Once done, the output is sent to Elasticsearch
3. Elasticsearch stores the document it got and the document is indexed for quick future access
4. The user creates visualizations in Kibana which based on the indexed data
5. The user creates a dashboard which composed out of the visualization created in the previous step

This is where data is stored and also where different processing takes place (e.g. when you search for a data).

While there can be multiple master nodes in reality only of them is the elected master node.

A node which responsible for parsing the data. In case you don't use logstash then this node can recieve data from beats and parse it, similarly to how it can be parsed in Logstash.

A Coordinating node responsible for routing requests out and in to the cluser (data nodes).

• Data is stored in an index
• The index is spread across the cluster using shards

"An index can be thought of as an optimized collection of documents and each document is a collection of fields, which are the key-value pairs that contain your data"

An index is split into shards and documents are hashed to a particular shard. Each shard may be on a different node in a cluster and each one of the shards is a self contained index.
This allows Elasticsearch to scale to an entire cluster of servers.

"An inverted index lists every unique word that appears in any document and identifies all of the documents each word occurs in."

Continuing with the comparison to SQL/NoSQL a Document in Elastic is a row in table in the case of SQL or a document in a collection in the case of NoSQL. As in NoSQL a Document is a JSON object which holds data on a unit in your app. What is this unit depends on the your app. If your app related to book then each document describes a book. If you are app is about shirts then each document is a shirt.

Red means some data is unavailable. Yellow can be caused by running single node cluster instead of multi-node.

"Each indexed field has a dedicated, optimized data structure. For example, text fields are stored in inverted indices, and numeric and geo fields are stored in BKD trees."

• _index
• _id
• _type

• You can optimize fields for partial matching
• You can define custom formats of known fields (e.g. date)
• You can perform language-specific analysis

In a network/cloud environment where failures can be expected any time, it is very useful and highly recommended to have a failover mechanism in case a shard/node somehow goes offline or disappears for whatever reason. To this end, Elasticsearch allows you to make one or more copies of your index’s shards into what are called replica shards, or replicas for short.

Term Frequency is how often a term appears in a given document and Document Frequency is how often a term appears in all documents. They both are used for determining the relevance of a term by calculating Term Frequency / Document Frequency.

"The index is actively being written to". More about the phases here

It creates customer index if it doesn't exists and adds a new document with the field name which is set to "John Dow". Also, if it's the first document it will get the ID 1.

1. If name value was different then it would update "name" to the new value
2. In any case, it bumps version field by one

Bulk API is used when you need to index multiple documents. For high number of documents it would be significantly faster to use rather than individual requests since there are less network roundtrips.

"In a filter context, a query clause answers the question “Does this document match this query clause?” The answer is a simple Yes or No — no scores are calculated. Filter context is mostly used for filtering structured data"

A small-scale architecture of elastic will consist of the elastic stack as it is. This means we will have beats, logstash, elastcsearch and kibana.
A production environment with large amounts of data can include some kind of buffering component (e.g. Reddis or RabbitMQ) and also security component such as Nginx.

• Input Plugins - how to collect data from different sources
• Filter Plugins - processing data
• Output Plugins - push data to different outputs/services/platforms

A logstash plugin which modifies information in one format and immerse it in another.

The raw data as it is stored in the index. You can search and filter it.

Total number of documents matching the search results. If not query used then simply the total number of documents.

"Visualize" is where you can create visual representations for your data (pie charts, graphs, ...)

Read here

False. One harvester harvests one file.

You can generate certificates with the provided elastic utils and change configuration to enable security using certificates model.

Another definition: "Systems that are physically separated, but logically connected"

• Network
• CPU
• Memory
• Disk

• Availability: Every request receives a response (it doesn't has to be the most recent data)
• Consistency: Every request receives a response with the latest/most recent data
• Partition tolerance: Even if some the data is lost/dropped, the system keeps running

What are the problems with the following design? How to improve it?

• Add another load balancer
• Use DNS A record for both load balancers
• Use message queue

It's an architecture in which data is and retrieved from a single, non-shared, source usually exclusively connected to one node as opposed to architectures where the request can get to one of many nodes and the data will be retrieved from one shared location (storage, memory, ...).

TODO: add more details!

"An explicitly and purposefully defined interface designed to be invoked over a network that enables software developers to get programmatic access to data and functionality within an organization in a controlled and comfortable way."

"An API specification provides a broad understanding of how an API behaves and how the API links with other APIs. It explains how the API functions and the results to expect when using the API"

"An API definition is similar to an API specification in that it provides an understanding of how an API is organized and how the API functions. But the API definition is aimed at machine consumption instead of human consumption of APIs."

Automation is the act of automating tasks to reduce human intervention or interaction in regards to IT technology and systems.
While automation focuses on a task level, Orchestration is the process of automating processes and/or workflows which consists of multiple tasks that usually across multiple systems.

• Authorization
• Logging
• Authentication
• Ordering
• Front-end
• Back-end ...

Data about data. Basically, it describes the type of information that an underlying data will hold.

I can't answer this for you :)

Domain Specific Language (DSLs) are used to create a customised language that represents the domain such that domain experts can easily interpret it.

Data serialization language used by many technologies today like Kubernetes, Ansible, etc.

True. Because YAML is superset of JSON.

someMultiLineString: |
  look mama
  I can write a multi-line string
  I love YAML

It's good for use cases like writing a shell script where each line of the script is a different command.

Wikipedia: "In computing, firmware is a specific class of computer software that provides the low-level control for a device's specific hardware. Firmware, such as the BIOS of a personal computer, may contain basic functions of a device, and may provide hardware abstraction services to higher-level software such as operating systems."

• Resolve host by request to DNS resolver
• Client SYN
• Server SYN+ACK
• Client SYN
• HTTP request
• HTTP response

False. It doesn't maintain state for incoming request.

• Request line - request type
• Headers - content info like length, enconding, etc.
• Body (not always included)

• GET
• POST
• HEAD
• PUT
• DELETE
• CONNECT
• OPTIONS
• TRACE

• 1xx - informational
• 2xx - Success
• 3xx - Redirect
• 4xx - Error, client fault
• 5xx - Error, server fault

TODO: explain what is actually a Cookie

The server didn't receive a response from another server it communicates with in a timely manner.

Wikipedia: "The X-Forwarded-For (XFF) HTTP header field is a common method for identifying the originating IP address of a client connecting to a web server through an HTTP proxy or load balancer."

A load balancer accepts (or denies) incoming network traffic from a client, and based on some criteria (application related, network, etc.) it distributes those communications out to servers (at least one).

• Scalability - using a load balancer, you can possibly add more servers in the backend to handle more requests/traffic from the clients, as opposed to using one server.
• Redundancy - if one server in the backend dies, the load balancer will keep forwarding the traffic/requests to the second server so users won't even notice one of the servers in the backend is down.

• Round Robin
• Weighted Round Robin
• Least Connection
• Weighted Least Connection
• Resource Based
• Fixed Weighting
• Weighted Response Time
• Source IP Hash
• URL Hash

• A simple round robin algorithm knows nothing about the load and the spec of each server it forwards the requests to. It is possible, that multiple heavy workloads requests will get to the same server while other servers will got only lightweight requests which will result in one server doing most of the work, maybe even crashing at some point because it unable to handle all the heavy workloads requests by its own.
• Each request from the client creates a whole new session. This might be a problem for certain scenarios where you would like to perform multiple operations where the server has to know about the result of operation so basically, being sort of aware of the history it has with the client. In round robin, first request might hit server X, while second request might hit server Y and ask to continue processing the data that was processed on server X already.

L4 and L7

Yes, you can use DNS for performing load balancing.

• Can cause uneven load on instance (since requests routed to the same instances) Pros:
• Ensures in-proc sessions are not lost when a new request is created

You would like to make sure the user doesn't lose the current session data.

Cookies. There are application based cookies and duration based cookies.

• Generated by the application and/or the load balancer
• Usually allows to include custom data

• Generated by the load balancer
• Session is not sticky anymore once the duration elapsed

In Copyleft, any derivative work must use the same licensing while in permissive licensing there are no such condition. GPL-3 is an example of copyleft license while BSD is an example of permissive license.

CPU cache. Source

SSH HTTP DHCP DNS ...

https://idiallo.com/blog/c10k-2016

• File
• Block
• Object

• Data is divided to self-contained objects
• Objects can contain metadata

• Usually with object storage, you pay for what you use as opposed to other storage types where you pay for the storage space you allocate
• Scalable storage: Object storage mostly based on a model where what you use, is what you get and you can add storage as need Cons:
• Usually performs slower than other types of storage
• No granular modification: to change an object, you have re-create it

• File Storage used for storing data in files, in a hierarchical sturcture
• Some of the devices for file storage: hard drive, flash drive, cloud-based file storage
• Files usually organized in directories

• Users have full control of their own files and can run variety of operations on the files: delete, read, write and move.
• Security mechanism allows for users to have a better control at things such as file locking

Local filesystem Dropbox Google Drive

This is a great way to figure how managers deal with unplanned work, and how good they are at setting expectations with projects.

This can give you insights in some of the cool projects a company is working on, and if you would enjoy working on projects like these. This is also a good way to see if the managers are allowing employees to learn and grow with projects outside of the normal work you'd do.

For Example, if they mention they have problem X, and you've solved that in the past, you can show how you'd be able to mitigate that problem.

Not only this will tell you what is expected from you, it will also provide big hint on the type of work you are going to do in the first months of your job.

• Data Storage: system to store data in tables
• SQL: programming language to manage relational databases
• Data Definition Language: a standard syntax to create, alter and delete tables

ACID stands for Atomicity, Consistency, Isolation, Durability. In order to be ACID compliant, the database must meet each of the four criteria

Atomicity - When a change occurs to the database, it should either succeed or fail as a whole.

For example, if you were to update a table, the update should completely execute. If it only partially executes, the update is considered failed as a whole, and will not go through - the DB will revert back to it's original state before the update occurred. It should also be mentioned that Atomicity ensures that each transaction is completed as it's own stand alone "unit" - if any part fails, the whole statement fails.

Consistency - any change made to the database should bring it from one valid state into the next.

For example, if you make a change to the DB, it shouldn't corrupt it. Consistency is upheld by checks and constraints that are pre-defined in the DB. For example, if you tried to change a value from a string to an int when the column should be of datatype string, a consistent DB would not allow this transaction to go through, and the action would not be executed

Isolation - this ensures that a database will never be seen "mid-update" - as multiple transactions are running at the same time, it should still leave the DB in the same state as if the transactions were being run sequentially.

For example, let's say that 20 other people were making changes to the database at the same time. At the time you executed your query, 15 of the 20 changes had gone through, but 5 were still in progress. You should only see the 15 changes that had completed - you wouldn't see the database mid-update as the change goes through.

Durability - Once a change is committed, it will remain committed regardless of what happens (power failure, system crash, etc.). This means that all completed transactions must be recorded in non-volatile memory.

Note that SQL is by nature ACID compliant. Certain NoSQL DB's can be ACID compliant depending on how they operate, but as a general rule of thumb, NoSQL DB's are not considered ACID compliant

Are you able to explain what is it good for?

Not much information provided as to why it became a bottleneck and what is current architecture, so one general approach could be
to reduce the load on your database by moving frequently-accessed data to in-memory structure.

Connection Pool is a cache of database connections and the reason it's used is to avoid an overhead of establishing a connection for every query done to a database.

A connection leak is a situation where database connection isn't closed after being created and is no longer needed.

• Query for running queries and cancel the irrelevant queries
• Check for connection leaks (query for running connections and include their IP)
• Check for table locks and kill irrelevant locking sessions

"A data warehouse is a subject-oriented, integrated, time-variant and non-volatile collection of data in support of organisation's decision-making process"

A database index is a data structure that improves the speed of operations in a table. Indexes can be created using one or more columns, providing the basis for both rapid random lookups and efficient ordering of access to records.

Data that is used multiple times in a database should be stored once and referenced with a foreign key.
This has the clear benefit of ease of maintenance where you need to change a value only in a single place to change it everywhere.

Primary Key: each row in every table should a unique identifier that represents the row.
Foreign Key: a reference to another table's primary key. This allows you to join table together to retrieve all the information you need without duplicating data.

• Primary data table: main data you care about
• Details table: includes a foreign key and has one to many relationship
• Lookup values table: can be one table per lookup or a table containing all the lookups and has one to many relationship
• Multi reference table

• Database as code
• Database abstraction
• Encapsulates SQL complexity
• Enables code review process
• Enables usage as a native OOP structure

Wikipedia: "In the context of SQL, data definition or data description language (DDL) is a syntax for creating and modifying database objects such as tables, indices, and users."

Bonus: extract the last word of each line

CDN (Content Delivery Network) responsible for distributing content geographically. Part of it, is what is known as edge locations, aka cache proxies, that allows users to get their content quickly due to cache features and geographical distribution.

In single CDN, the whole content is originated from content delivery network.
In multi-CDN, content is distributed across multiple different CDNs, each might be on a completely different provider/cloud.

• Resiliency: Relying on one CDN means no redundancy. With multiple CDNs you don't need to worry about your CDN being down
• Flexibility in Costs: Using one CDN enforces you to specific rates of that CDN. With multiple CDNs you can take into consideration using less expensive CDNs to deliver the content.
• Performance: With Multi-CDN there is bigger potential in choosing better locations which more close to the client asking the content
• Scale: With multiple CDNs, you can scale services to support more extreme conditions

• Not suitable for frequent code changes and the ability to deploy new features
• Not designed for today's infrastructure (like public clouds)
• Scaling a team to work monolithic architecture is more challenging

• Each of the services individually fail without escalating into an application-wide outage.
• Each service can be developed and maintained by a separate team and this team can choose its own tools and coding language

This article provides a great explanation.

The ability easily grow in size and capacity based on demand and usage.

The ability to grow but also to reduce based on what is required

High Availability - Being able to access a resource (in some use cases, using different platforms)

wintellect.com: "High availability, simply put, is eliminating single points of failure and disaster recovery is the process of getting a system back to an operational state when a system is rendered inoperative. In essence, disaster recovery picks up when high availability fails, so HA first."

Vertical Scaling is the process of adding resources to increase power of existing servers. For example, adding more CPUs, adding more RAM, etc.

With vertical scaling alone, the component still remains a single point of failure. In addition, it has hardware limit where if you don't have more resources, you might not be able to scale vertically.

Databases, cache. It's common mostly for non-distributed systems.

Horizontal Scaling is the process of adding more resources that will be able handle requests as one unit

A load balancer. You can add more resources, but if you would like them to be part of the process, you have to serve them the requests/responses. Also, data inconsistency is a concern with horizontal scaling.

Take a look here

You can find a list here

Read about it here

roll-back & roll-forward cut over dress rehearsals DNS redirection

A central processing unit (CPU) performs basic arithmetic, logic, controlling, and input/output (I/O) operations specified by the instructions in the program. This contrasts with external components such as main memory and I/O circuitry, and specialized processors such as graphics processing units (GPUs).

RAM (Random Access Memory) is the hardware in a computing device where the operating system (OS), application programs and data in current use are kept so they can be quickly reached by the device's processor. RAM is the main memory in a computer. It is much faster to read from and write to than other kinds of storage, such as a hard disk drive (HDD), solid-state drive (SSD) or optical drive.

An embedded system is a computer system - a combination of a computer processor, computer memory, and input/output peripheral devices—that has a dedicated function within a larger mechanical or electronic system. It is embedded as part of a complete device often including electrical or electronic hardware and mechanical parts.

Raspberry Pi

• Volume: Extremely large volumes of data
• Velocity: Real time, batch, streams of data
• Variety: Various forms of data, structured, semi-structured and unstructured
• Veracity or Variability: Inconsistent, sometimes inaccurate, varying data

DataOps seeks to reduce the end-to-end cycle time of data analytics, from the origin of ideas to the literal creation of charts, graphs and models that create value. DataOps combines Agile development, DevOps and statistical process controls and applies them to data analytics.

"Data architecture is the process of standardizing how organizations collect, store, transform, distribute, and use data. The goal is to deliver relevant data to people who need it, when they need it, and help them make sense of it."

• Structured - data that has defined format and length (e.g. numbers, words)
• Semi-structured - Doesn't conform to a specific format but is self-describing (e.g. XML, SWIFT)
• Unstructured - does not follow a specific format (e.g. images, test messages)

Wikipedia's explanation on Data Warehouse Amazon's explanation on Data Warehouse

Data Lake - Wikipedia

Apache Hadoop - Wikipedia

Responsible for managing the compute resources in clusters and scheduling users' applications

A programming model for large-scale data processing

• Distributed file system providing high aggregate bandwidth across the cluster.
• For a user it looks like a regular file system structure but behind the scenes it's distributed across multiple machines in a cluster
• Typical file size is TB and it can scale and supports millions of files
• It's fault tolerant which means it provides automatic recovery from faults
• It's best suited for running long batch operations rather than live analysis

• Master-slave architecture
• Namenode - master, Datanodes - slaves
• Files split into blocks
• Blocks stored on datanodes
• Namenode controls all metadata

• Object (RGW)
• Block (RBD)
• File (CephFS)

• Reliable Autonomic Distributed Object Storage
• Provides low-level data object storage service
• Strong Consistency
• Simplifies design and implementation of higher layers (block, file, object)

• Monitor
  • Central authority for authentication, data placement, policy
  • Coordination point for all other cluster components
  • Protect critical cluster state with Paxos
• Manager
  • Aggregates real-time metrics (throughput, disk usage, etc.)
  • Host for pluggable management functions
  • 1 active, 1+ standby per cluster
• OSD (Object Storage Daemon)
  • Stores data on an HDD or SSD
  • Services client IO requests

In general, Packer automates machine images creation. It allows you to focus on configuration prior to deployment while making the images. This allows you start the instances much faster in most cases.

1. Deployment Speed - you configure once prior to deployment instead of configuring every time you deploy. This allows you to start instances/services much quicker.
2. More immutable infrastructure - with configuration->deployment it's not likely to have very different deployments since most of the configuration is done prior to the deployment. Issues like dependencies errors are handled/discovered prior to deployment in this model.