1.
01

2.
02

3.
Yamlware
Basically, any (configuration|infrastructure|deployment)-as-code
software nowadays.
“
03

4.
04

5.
05

6.
YAML: Good
or bad?
06

7.
YAML is
great!
07

8.
For computers
YAML is a human friendly data serialization standard for all
programming languages.
“
08

9.
For humans
YAML is a human friendly data serialization standard for all
programming languages.
“
09

10.
10

11.
11

12.
YAML is
awful!
12

13.
13

14.
14

15.
#NoYaml
15

16.
Did you know?
!!python/object/apply:os.system
args: ['ls /']
01.
02.
16

17.
Duplicate keys
- name: A
name: B
01.
02.
17

18.
Duplicate keys
18

19.
Indentation
- name: nano
apt:
name: nano
01.
02.
03.
19

20.
Base 60
Using ":" allows expressing integers in base 60, which is convenient
for time and angle values.
“
20

21.
Port mappings
25:25 = 25*60 + 25*3600 = 91500
53:53 = ... = 193980
•
•
21

22.
Anchors and references
- &sometask
name: "Some Task"
some_module: with a lot of parameters
with_items: list_A
- <<: *sometask
name: "Some OTHER Task"
with_items: list_B
01.
02.
03.
04.
05.06.
07.
08.
22

23.
Boolean
y|Y|yes|Yes|YES|
n|N|no|No|NO|
true|True|TRUE|
false|False|FALSE|
on|On|ON|
off|Off|OFF
01.
02.
03.
04.
05.
06.
23

24.
24

25.
Content is
trunca
25

26.
Yaml is not
Turing-
complete!26

27.
Ok, that's
unfair!
27

28.
Of course, it's not!
It's not a computational language, it's a data language
List of maps of lists of maps of maps of lists...
•
•
28

29.
Is JSON
better?
29

30.
No comments
30

31.
XML?
Anyone?
31

32.
32

33.
Stop whining
33

34.
Yaml is a tool!
34

35.
Industry needs a data language
We need a simple language, to declare our intent, describe
configuration, manage state, share serialized data.
35

36.
Industry
needs it,
industry
creates it36

37.
JSONNET
37

38.
JSON5
38

39.
TOML
39

40.
DHALL
40

41.
YTT
41

42.
HCL/HCL2
42

43.
Standards, standards,...
43

44.
How about a real language?
44

45.
Declarations, not actions
Yaml is good for defining serialized data and declared state
But it does not support constructs for variables, functions, deep
references
And it is not so good for imperative/sequential scripts (I'm looking at
you Ansible)
?
•
•
•
45

46.
46

47.
The machine
47

48.
48

49.
The machine
Yaml (or any similar language) is just the declaration of WHAT
Heavy lifting of HOW is done by the machinery
•
•
49

50.
50

51.
51

52.
New version
They just released v1.16 this month
52

53.
U.S.E.
$ use version
1.16 (Orbit)
01.
02.
53

54.
CLI
$ use apply solar_system.yml01.
54

55.
55

56.
56

57.
Let's play!
57

58.
58

59.
Pluto's status
kind: Planet
metadata:
type: planet
name: Pluto
01.
02.
03.
04.
59

60.
Pluto's status
kind: Planet
metadata:
type: dwarf_planet
name: Pluto
01.
02.
03.
04.
60

61.
Pluto's status
$ use apply solar_system.yml
use: Modifying Pluto...
use: Pluto's metadata modified in 10ns
fume: libraries on Earth have been updated
to reflect changes in Pluto's status
01.
02.
03.
04.
05.
61

62.
62

63.
Comet
kind: Comet
metadata:
name: MyComet
replicas: 1
speed: 1000km/s
01.
02.
03.
04.
05.
63

64.
Let's check it
$ use apply solar_system.yml
use: Creating "MyComet"...
use: MyCommet is created
fume: WARNING: Unknown object is aproaching planet "Earth"
01.
02.
03.
04.
64

65.
Let's remove the code
$ use apply solar_system.yml
use: all is up-to-date. no changes have been applied.
fume: WARNING: Unknown object is aproaching planet "Earth"
01.
02.
03.
65

66.
Compensating actions
$ use delete comet/MyComet
use: Comet "MyComet" has been destroyed
01.
02.
66

67.
67

68.
That's no
moon
68

69.
That's no moon
kind: Planet
metadata:
type: planet
name: Earth
moons:
- name: Moon
- name: Moon2
01.
02.
03.
04.
05.
06.
07.
69

70.
Go! Go! Go!
$ use apply solar_system.yml
use: moon "Moon" already exists. Skipped.
use: Creating moon "Moon2"...
use: Done in 2 years 23 days.
use: Moon2 is up and running.
01.
02.
03.
04.
05.
70

71.
Wait! What?
fume: WARNING: Earth: unexpected floods in
Sri-Lanka due to irregular tides patterns
fume: WARNING: Earth: rotation slowed down by 0.3%.
Risk of seasonal shift!
01.
02.
03.
04.
71

72.
SOL-101: Bug report
Earth is overpopulated. We need to scale it up to house more
biological forms.
72

73.
Bug fix
kind: Planet
metadata:
type: planet
name: Earth
dimensions:
radius: 8000km
01.
02.
03.
04.
05.
06.
73

74.
Push it
$ git add solar_system.yml
$ git commit -m "SOL-101: increased Earth radius "
1 file changed, 1 insertions(+), 0 deletions(-)
$ git push
b64049e..ed99f47 master -> master
01.
02.
03.
04.
05.
74

75.
No way!
jobs: Starting job for commit b64049e
jobs: Running script
jobs: use apply solar_system.yml
use: Destroying planet Earth...
use: Still destroying...
fume: FATAL: Earth is being destroyed...
01.
02.
03.
04.
05.
06.
75

76.
Backup
kind: PlanetBackup
target:
name: Earth
schedule:
cron: * * * * *
01.
02.
03.
04.
05.
76

77.
10 days later...
fume: There is no space left in space.01.
77

78.
Real life state machines
Kubernetes
Terraform
CloudFormation
Serverless.com
Ansible
Puppet
etc.
•
•
•
•
•
•
•
78

79.
Idempotance
79

80.
Idempotance
80

81.
Idempotent
#!/bin/bash
apt-get -y install nano
01.
02.
81

82.
Not idempotent
#!/bin/bash
apt-get -y install nano
useradd andrey
useradd: user 'andrey' already exists
01.
02.
03.
04.05.
82

83.
More code
#!/bin/bash
apt-get -y install nano
if ! id -u andrey > /dev/null 2>&1; then
useradd andrey
fi
01.
02.
03.
04.
05.
83

84.
Declarative catalogue
- user:
name: andrey
shell: /bin/zsh
- package:
name: nano
version: 1.23.1-1
01.
02.
03.
04.
05.
06.
84

85.
Readability!
85

86.
Pipeline-
integratable
86

87.
Abstraction
87

88.
Resource
graph
88

89.
Parallel
execution
89

90.
Automatic
ordering
90

91.
Modularity
91

92.
Data query
- package:
name: nano-${os.version}
version: 1.23.1-1
01.
02.
03.
92

93.
Resource deletion
- user:
name: andrey
shell: /bin/zsh
# - package:
# name: nano
# version: 1.23.1-1
01.
02.
03.
04.
05.
06.
93

94.
Attribute management
Write-once
Updatable, applied right-away
Updatable, requires extra action to take effect
Updatable, forces resource recreation
•
•
•
•
94

95.
Resource
recreation
95

96.
Resource
protection
96

97.
Dry-run
97

98.
Compensating
Compensating actions may sometimes be required
Not every state change is allowed
Any software has bugs!
•
•
•
98

99.
State verification
Deployment-time
Run-time
•
•
99

100.
Module-level
dependencies
100

101.
Resource-
level
dependencies101

102.
SM-index
idempotence maturity
declarativeness maturity (readiblity)
data query richness
execution engine maturity (parallel execution, auto-ordering of
dependencies, deletion of obsolete objects)
modularity maturity
documentation maturity
•
•
•
•
•
•
102

103.
103

104.
104

105.
Ansible
machine
105

106.
106

107.
Ansible machine
Signs of state management, idempotance is not always achievable.
Can be parallel only on node level, not resource (task) level.
No model for resource dependencies.
Imperative logic elements implemented in a declarative, but hardly
readable way ( register , set_fact , when ).
•
•
•
•
107

108.
Kubernetes
machine
108

109.
109

110.
Kubernetes machine
api-server will accept any valid object definition.
Will it be created and running? Eventually, it should.
Run-time state validation.
Not all attributes are changeable.
Custom resource definitions (+ operators).
Queries and dependencies only through labels.
•
•
•
•
•
•
110

111.
kubectl wait
111

112.
Helm
112

113.
Terraform
machine
113

114.
114

115.
Terraform machine
Terraform remembers previous state.
Deployment time validation, no run-time validation.
Not YAML, rich expression language.
Powerful query capabilities.
•
•
•
•
115

116.
Final words
116

117.
Side effects are real
Know the tool
Know which resources you are managing
Know the side effects
•
•
•
117

118.
Tools
There are tools for tools: linters, IDE plugins, schema validators.
118

119.
Everything is a spectrum
Avoid dichotomy (bad/good, persistent/volatlie etc.), use a spectrum to
define your relationship with a tool/language.
119

120.
Keep it sane
If you need a complex logic, use a %&?# programming language or a
better declarative language with functions and variables.
Infrastructure-as-code does not mean you should use the same data
language all over the place.
Do not use YAML for imperative logic. Do not let the history repeating.
•
•
•
120

121.
Keep it simple
121

122.
Thank you!
122

123.
123

124.
124

125.
References
125

126.
Whining
https://arp242.net/yaml-config.html
https://noyaml.com
https://boyter.org/posts/an-informal-survey-of-10-million-github-
bitbucket-gitlab-projects/#yaml-or-yml
https://www.darkcoding.net/software/a-developer-goes-to-a-devops-
conference/
•
•
•
•
126

127.
Languages
https://jsonnet.org/
https://json5.org/
https://dhall-lang.org/
https://get-ytt.io/
https://github.com/bazelbuild/starlark
https://github.com/hashicorp/hcl2/blob/master/hcl/hclsyntax/spec.md
https://github.com/toml-lang/toml#toml
https://www.openpolicyagent.org/docs/latest/policy-language/
https://en.wikipedia.org/wiki/Datalog
•
•
•
•
•
•
•
•
• 127

128.
Tips
https://www.redhat.com/sysadmin/yaml-tips
https://networkgenomics.com/ansible/
•
•
128

129.
Kubernetes
https://www.infoq.com/articles/kubernetes-workloads-serverless-era/•
129