43
In e na ional Jou nal o Ad ance and Applied Resea ch
www.ijaa .co.in
ISSN – 2347-7075
Impac Fac o – 8.141
Pee Re iewed
Bi-Mon hly
Vol. 6 No. 38
Sep embe - Oc obe - 2025
Scalabili y and Rollback E iciency o Kube ne es Deploymen Pa e ns: A
S udy o Blue-G een s. Cana y App oaches
Ami K. Mogal1 & Vaibha P. Sonaje2
1Depa men o Compu e Science and Applica ion,
MVP Samaj’s Comme ce, Managemen and Compu e Science (CMCS) College, Nashik,
Maha ash a, India.
2Depa men o Compu e Science and Applica ion, School o Compu e Science and
Enginee ing, Sandip Uni e si y, Nashik, Maha ash a, India
Co esponding Au ho – Ami K. Mogal
DOI - 10.5281/zenodo.17309883
Abs ac :
In cloud-na i e so wa e de elopmen , con inuous deploymen s a egies signi ican ly in luence
applica ion a ailabili y, eliabili y, and main ainabili y. This s udy p esen s a compa a i e analysis o
wo widely adop ed Kube ne es deploymen pa e ns—Blue-G een and Cana y— ocusing on hei
scalabili y, ollback e iciency, and esou ce u iliza ion in mic ose ices-based web applica ions. Using
a con olled Kube ne es en i onmen , a ic simula ions we e execu ed ia K6 o eplica e eal-wo ld
load scena ios, including linea amp-ups, bu s loads, and sus ained use concu ency. Key
pe o mance me ics such as pod s a up ime, eques la ency (P50, P90, P99), CPU/memo y
u iliza ion, and ollback du a ion we e collec ed and analyzed using P ome heus and G a ana
dashboa ds. Resul s show ha while Blue-G een deploymen s o e as e ollback and simple e sion
con ol, Cana y deploymen s p o ide ine a ic con ol and g ea e aul isola ion du ing inc emen al
eleases. The indings highligh c i ical ade-o s in deploymen s a egy selec ion and p o ide
ope a ional insigh s o De Ops eams seeking o op imize eliabili y and se ice con inui y in
Kube ne es clus e s. The s udy con ibu es o deploymen au oma ion bes p ac ices and suppo s
in o med decision-making o scalable, esilien mic ose ice deli e y pipelines.
Keywo ds: Kube ne es, Mic ose ices, Blue-G een Deploymen , Cana y Deploymen , Con inuous
Deli e y.
In oduc ion:
The e olu ion o so wa e deli e y has
been p o oundly in luenced by he ise o
mic ose ices a chi ec u e and he ad en o
con aine o ches a ion pla o ms like
Kube ne es. In oday’s as -paced de elopmen
landscape, con inuous in eg a ion and
con inuous deploymen (CI/CD) p ac ices a e
essen ial o achie ing apid, eliable so wa e
deli e y. Wi hin his con ex , deploymen
s a egies play a c i ical ole in ensu ing ha
new so wa e e sions can be olled ou
e icien ly, wi hou dis up ing use expe ience
o comp omising sys em s abili y. Two o he
mos widely used deploymen pa e ns in
Kube ne es-based en i onmen s a e Blue-
G een deploymen s and Cana y deploymen s,
each o e ing unique ad an ages and ade-o s
in e ms o scalabili y, aul ole ance, and
ollback capabili ies.
Blue-G een deploymen is a well-
es ablished echnique in which wo iden ical
en i onmen s— e e ed o as "blue" and
"g een"—a e main ained. The li e
en i onmen se es he cu en p oduc ion
a ic, while he new e sion is deployed o
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
44
he idle en i onmen . A e success ul
alida ion, he ou e o load balance swi ches
a ic o he new e sion, allowing ins an
ollback by e e ing a ic o he p e ious
en i onmen in case o ailu e. This s a egy
p o ides a high deg ee o con ol and
immedia e ollback capabili y bu comes a he
cos o duplica ing in as uc u e and po en ial
unde u iliza ion o esou ces. On he o he
hand, Cana y deploymen in ol es eleasing
he new e sion inc emen ally o a subse o
use s o a ic while he majo i y con inues o
in e ac wi h he s able elease. This g adual
ollou enables eal- ime moni o ing o
pe o mance and e o a es, allowing eams o
de ec and espond o issues be o e he ull
deploymen . While his me hod enhances aul
isola ion and educes he isk o ull-scale
ailu es, i in oduces complexi y in a ic
managemen and equi es obus obse abili y
o be e ec i e. Kube ne es, as a con aine
o ches a ion pla o m, p o ides na i e suppo
and ex ensibili y o implemen ing hese
deploymen s a egies h ough se ices,
ing ess con olle s, and cus om esou ce
de ini ions (CRDs). Howe e , he p ac ical
e iciency o Blue-G een and Cana y
deploymen s in Kube ne es en i onmen s—
pa icula ly in e ms o scalabili y unde high
use load and ollback pe o mance du ing
ailu e scena ios— emains an a ea ha
demands empi ical alida ion. As
o ganiza ions adop mic ose ices a scale,
unde s anding he ope a ional impac o hese
deploymen pa e ns is c ucial o main aining
se ice a ailabili y and pe o mance.
This s udy seeks o add ess his gap by
conduc ing a comp ehensi e compa a i e
analysis o Blue-G een and Cana y
deploymen app oaches in Kube ne es,
ocusing speci ically on hei scalabili y and
ollback e iciency. A se ies o con olled
expe imen s we e conduc ed using a
mic ose ices-based web applica ion deployed
in Kube ne es, subjec ed o simula ed load
pa e ns gene a ed using K6, a mode n
pe o mance es ing ool. Me ics such as CPU
and memo y u iliza ion, pod s a up ime,
eques la ency ac oss pe cen iles (P50, P90,
P99), ollback du a ion, and au oscaling
beha io we e cap u ed h ough P ome heus
and G a ana moni o ing s acks. The p ima y
objec i e o his esea ch is o e alua e which
deploymen pa e n o e s supe io
pe o mance and esilience unde a ious
s ess condi ions, and how hese app oaches
can be op imized o mee he demands o eal-
ime, la ge-scale web applica ions. The
indings o his s udy p o ide aluable insigh s
o De Ops enginee s, cloud a chi ec s, and
so wa e eams seeking o imp o e he
eliabili y, e iciency, and aul eco e y o
hei deploymen pipelines in Kube ne es
en i onmen s. This s udy aims o compa e he
pe o mance o Blue-G een and Cana y
deploymen s a egies in Kube ne es
en i onmen s unde a ious a ic loads,
including linea , bu s , and sus ained a ic.
Addi ionally, i e alua es he ollback
e iciency o bo h s a egies by simula ing
con olled ailu e scena ios o assess hei
eliabili y and esponsi eness. Ul ima ely, he
esea ch seeks o de e mine which deploymen
s a egy o e s supe io scalabili y, lowe
la ency, and mo e esou ce-e icien ollback
beha io , p o iding aluable insigh s o
op imizing Kube ne es clus e pe o mance
and eliabili y. By sys ema ically analyzing
deploymen ou comes unde consis en
wo kload scena ios, his esea ch con ibu es
o he ongoing discou se on bes p ac ices in
cloud-na i e deploymen au oma ion, o e ing
a da a-d i en ounda ion o making s a egic
decisions abou deploymen me hodologies in
p oduc ion Kube ne es en i onmen s
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
45
Rela ed Wo k:
The ad ancemen o mic ose ices
a chi ec u e and he widesp ead adop ion o
Kube ne es ha e b ough a pa adigm shi in
applica ion deploymen s a egies. Among he
mos p ominen me hods o achie ing
con inuous deploymen and ze o-down ime
deli e y a e Blue-G een and Cana y
deploymen s. These echniques ha e been
ex ensi ely explo ed in bo h indus y and
academia, pa icula ly in ela ion o
pe o mance, scalabili y, ollback e iciency,
and isk mi iga ion. James and Gideon (2024)
conduc ed a comp ehensi e e alua ion o
Blue-G een and Cana y deploymen s ac oss
mul iple axes, including scalabili y, aul
ole ance, and moni o ing o e head. Thei
s udy highligh ed ha Blue-G een
deploymen s o e a mo e s aigh o wa d
ollback mechanism by swi ching a ic
be ween p oduc ion and s aging en i onmen s
bu come a he cos o duplica ing esou ces.
In con as , Cana y deploymen s p o ide
g anula a ic con ol and eal- ime e o
de ec ion, albei equi ing obus obse abili y
o ensu e sa e y and e ec i eness du ing
p og essi e ollou s.
In en e p ise-le el Kube ne es
en i onmen s, P abu (2024) emphasized he
a chi ec u al di e ences and implica ions o
using Blue-G een e sus Cana y deploymen s.
He ound ha Blue-G een app oaches simpli y
deploymen o s a e ul applica ions due o
ixed ou ing bu end o in oduce
ine iciencies when applied o dynamic
mic ose ices, especially unde high
concu ency. Con e sely, Cana y deploymen s
demons a ed supe io adap abili y in
au oscaling scena ios, hanks o Kube ne es-
na i e in eg a ions like cus om me ics APIs
and se ice mesh ou ing (e.g., Is io). Vangala
(2025) compa ed he wo s a egies
speci ically wi hin he con ex o De Ops
wo k lows. His s udy concluded ha Cana y
deploymen s ou pe o m Blue-G een in e ms
o sys em esilience and la ency managemen
du ing high-load es ing phases, especially
when ollback needs o be pa ial o selec i e.
Blue-G een’s ins an ollback was ound o be
mo e e icien only in con olled, low- a ic
scena ios whe e ull-sys em swi ches a e
iable.
Addi ional wo k by Idowu (2024)
e alua ed ollback ime, e o a es, and
sys em eco e y ac oss Blue-G een and
Cana y s a egies using Kube ne es and Is io.
His indings aligned wi h p io esea ch,
ein o cing ha Cana y deploymen allows o
ea ly anomaly de ec ion and inc emen al
ailu e isola ion, which is especially c i ical in
la ge-scale CI/CD pipelines whe e down ime
can ha e cascading e ec s. His s udy also
emphasized he impo ance o obse abili y
amewo ks like P ome heus and G a ana in
ensu ing sa e cana y eleases. Deploymen
ollback mechanisms in complex pipelines
we e u he in es iga ed by William and
Me cy (2025), who explo ed how ollback
logic is handled ac oss di e en ools,
including Kube ne es, A goCD, and
Spinnake . They no ed ha while bo h Blue-
G een and Cana y deploymen s suppo
ollback p ocedu es, he unde lying igge and
execu ion mechanisms di e signi ican ly. In
Blue-G een, ollback is de e minis ic and
swi ch-based; in Cana y, ollback is o en
e en -d i en and in luenced by eal- ime
pe o mance eleme y.
In e ms o scalabili y, Rakshi and
Bane jee (2024) pe o med benchma king
expe imen s on Kube ne es clus e s unde
a iable loads using bo h deploymen models.
They concluded ha Cana y deploymen s
scale mo e p edic ably when combined wi h
Ho izon al Pod Au oscale s (HPA) and cus om
me ics, while Blue-G een deploymen s
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
46
exhibi ed delayed esou ce s abiliza ion due o
ab up a ic shi ing. A signi ican
con ibu ion om Reid and James (2025)
in ol ed au oma ing bo h Blue-G een and
Cana y deploymen pipelines using Te a o m.
Thei wo k demons a ed ha in as uc u e-
as-code ools could educe human e o and
p omo e epea abili y in deploymen s a egies.
They ound ha Cana y models bene i ed
mo e om dynamic empla ing and modula
in as uc u e due o hei p og essi e na u e
and need o g anula con ol.
In he ealm o CI/CD op imiza ion,
Amgo hu (2024) ocused on in eg a ing cana y
and blue-g een s a egies in o CI pipelines
using Jenkins and Gi Lab CI. His expe imen s
showed ha Cana y deploymen s esul ed in
ewe p oduc ion ollbacks when in eg a ed
wi h pe o mance ale ing and anomaly
de ec ion sys ems, hus educing o e all
MTTR (mean ime o eco e y). Finally, he
wo k by Sun-Rise (2025) in a p oduc ion-
g ade si e eliabili y enginee ing (SRE)
con ex emphasized ha he combina ion o
p og essi e deploymen models wi h
au oma ed ollback mechanisms p o ides he
mos obus deploymen eliabili y. His
analysis showed ha Blue-G een is op imal o
con olled eleases, while Cana y excels in
high- eloci y, equen deploymen s.
Resea ch Me hodology:
This s udy adop s an expe imen al
esea ch design o e alua e and compa e he
scalabili y and ollback e iciency o wo
Kube ne es deploymen s a egies: Blue-G een
and Cana y. The objec i e is o simula e eal-
wo ld a ic loads, collec pe o mance
me ics, and analyze how each deploymen
s a egy esponds unde a ying condi ions
wi hin a con olled Kube ne es en i onmen .
Expe imen al En i onmen Se up:
Table 1. En i onmen al Se up o Expe imen .
Componen
Desc ip ion
Pla o m
Kube ne es 1.24 unning on Minikube and AWS EKS ( o
scalabili y alida ion)
Applica ion
Web-based mic ose ice (Node.js backend + Reac on end
+ MongoDB)
CI/CD Tool
Gi Lab CI wi h in eg a ion o A goCD and Helm
T a ic
Gene a o
K6 – o simula ing ealis ic load pa e ns
Obse abili y
Tools
P ome heus, G a ana, and Loki ( o logs, me ics, and ale s)
Se ice Mesh
Is io ( o a ic ou ing and Cana y ollou con ol)
Deploymen Pa e ns Unde S udy:
Blue-G een deploymen in ol es
deploying a new e sion o an applica ion in
pa allel wi h he exis ing e sion. Once he
new e sion is alida ed, a ic is shi ed
en i ely o he new e sion. One o he key
bene i s o his app oach is he abili y o
pe o m ins an ollbacks by simply swi ching
a ic back o he p e ious e sion i any
issues a ise. In con as , Cana y deploymen
ollows a mo e g adual app oach, whe e a new
e sion is olled ou o a small pe cen age o
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
47
a ic ini ially, such as 10% o 25%. The
deploymen is closely moni o ed o key
me ics like e o a es and la ency be o e
inc emen ally inc easing he exposu e o mo e
use s. I any p ede ined h esholds a e
b eached, he deploymen can be olled back
o ensu e minimal impac on use s. This
app oach allows o mo e con olled and isk-
a e se deploymen s.
Wo kload Simula ion S a egy:
T a ic pa e ns we e simula ed using
K6 o mimic eal-wo ld condi ions as shown in
Table 2 below
Table 2. Wo kload Simula ion o Expe imen .
Load Type
De ails
Linea Ramp-Up
G adual inc ease om 10 → 200 use s o e 10 minu es
Bu s Load
Sudden spike o 500 concu en use s
Sus ained Load
150+ concu en use s o e a 15-minu e window
Me ics Collec ed:
The s udy e alua ed se e al key
me ics ac oss mul iple ca ego ies, including:
1. Scalabili y: CPU u iliza ion, memo y
usage, pod s a up ime, and pod coun
unde Ho izon al Pod Au oscaling
(HPA).
2. Pe o mance: Reques la ency (P50,
P90, P99) and e o a es.
3. Rollback E iciency: Time o ollback
and numbe o ailed eques s du ing
ollback.
4. Sys em Th oughpu : Reques s pe
second (RPS) and success a e.
5. Cos E iciency: Resou ce u iliza ion
e sus wo kload se ed.
These me ics p o ide a
comp ehensi e unde s anding o he
pe o mance, scalabili y, and e iciency o he
deploymen s a egies.
Failu e Injec ion & Rollback Tes ing:
To e alua e ollback e iciency,
delibe a e aul s we e in oduced du ing
deploymen , including:
1. CPU sa u a ion using syn he ic
compu e-bound loads o simula e
esou ce exhaus ion.
2. Fo ced applica ion e o s h ough bad
con igu a ion o HTTP 500 e o s o
mimic eal-wo ld ailu es.
3. Moni o ing h esholds, such as e o
a es exceeding 2% o la ency abo e
500ms, we e se o igge ollbacks
au oma ically ia A goCD and Is io
ou ing policies.
These aul injec ion scena ios allowed
o assessing he esponsi eness and
e ec i eness o he ollback mechanisms in
bo h Blue-G een and Cana y deploymen
s a egies.
Da a Collec ion Tools and Logging:
The s udy u ilized se e al ools o
moni o ing and me ics collec ion:
1. P ome heus: Collec ed CPU, memo y,
and pod me ics.
2. G a ana: P o ided dashboa ds o
isualize me ic e olu ion o e ime.
3. Loki: Agg ega ed logs o de ec e o s
and ollback e en s.
4. K6 Ou pu JSON: Used o ex ac
de ailed la ency dis ibu ions and
h oughpu me ics.
These ools enabled comp ehensi e
moni o ing and analysis o he deploymen
s a egies' pe o mance.
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
48
Expe imen al P ocedu e:
The expe imen consis ed o se e al key s eps:
1. A baseline es was conduc ed by
deploying a s able e sion using bo h
Blue-G een and Cana y s a egies,
measu ing he idle o e head o each.
2. T a ic simula ion was hen
pe o med, in oducing loads
acco ding o a p ede ined scena io
ma ix.
3. A ollback scena io was igge ed
mid-deploymen , simula ing a aul
and measu ing he esponse ime and
sys em impac .
4. To ensu e consis ency, each scena io
was execu ed h ee imes, and he
a e age esul s we e used o analysis.
5. Resul s we e logged in JSON o ma
and G a ana snapsho s, acili a ing
de ailed pos -analysis and compa ison
o he wo deploymen s a egies.
Validi y & Reliabili y Measu es:
To ensu e a ai and eliable
compa ison, he expe imen u ilized iden ical
clus e con igu a ions and sys em baselines o
bo h deploymen s a egies. The ials we e
epea ed o minimize he impac o andom
luc ua ions, and he esul s we e alida ed
using li e me ics and s uc u ed logging. This
app oach enabled ep oducibili y and accu acy
in he indings, p o iding a solid ounda ion
o compa ing he pe o mance o Blue-G een
and Cana y deploymen s a egies
Resul , Analysis and Discussion:
This sec ion p esen s he expe imen al
indings compa ing Blue-G een and Cana y
deploymen s a egies in Kube ne es
en i onmen s, ocusing on key pe o mance
me ics such as scalabili y, esou ce
u iliza ion, ollback e iciency, and sys em
h oughpu unde simula ed wo kloads. The
esul s p o ide insigh s in o he s eng hs and
limi a ions o each s a egy, highligh ing hei
sui abili y o di e en use cases and
en i onmen s. By analyzing hese me ics, his
s udy aims o in o m bes p ac ices o
deploying applica ions in Kube ne es,
ul ima ely enhancing sys em eliabili y,
e iciency, and pe o mance.
Scalabili y Me ics: Ou esul s p o ide
insigh s in o he s eng hs and weaknesses o
each s a egy, in o ming decisions on op imal
deploymen app oaches in cloud-na i e
applica ions.
Table 3. CPU and Memo y U iliza ion
Time (min)
Clus e 1 (Blue-G een)
Clus e 2 (Cana y)
CPU Peak
74%
62%
Memo y Peak (MiB)
5571 MiB
4505 MiB
Cana y deploymen s demons a ed
supe io esou ce e iciency compa ed o Blue-
G een deploymen s, exhibi ing lowe CPU and
memo y usage due o hei p og essi e ollou
and g adual pod scaling. In con as , Blue-
G een deploymen s esul ed in ab up esou ce
p o isioning, leading o no able esou ce
spikes and ine icien pod u iliza ion du ing
a ic swi ching. This compa ison yields a key
insigh : Cana y deploymen o e s be e
esou ce elas ici y, e ec i ely a oiding sys em
sa u a ion du ing peak loads. By adop ing a
mo e g adual app oach o deploymen , Cana y
deploymen s minimize he s ain on sys em
esou ces, ensu ing mo e e icien and eliable
pe o mance unde a ying loads.
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
49
Table 4. Pod S a up Time and Au oscaling Responsi eness
Scena io
Blue-G een
Cana y
Cold S a (a g)
10.2s
7.4s
Au oscaling Response
5 pods in 10 min
6 pods in 6 min
When pai ed wi h Ho izon al Pod
Au oscaling (HPA) and me ic-based igge s,
Cana y deploymen enabled as e scaling
decisions, allowing o mo e agile esponses o
changing a ic condi ions. In con as , Blue-
G een deploymen equi ed a ull se o pods
o be eady be o e swi ching a ic, which
in oduced delays in eadiness and inc eased
s a up la ency. This highligh s a key
ad an age o Cana y deploymen : i s abili y o
p o ide ine con ol o e ollou eloci y and
scaling policies, making i mo e esponsi e o
dynamic a ic su ges. By le e aging his
lexibili y, Cana y deploymen can be e
adap o luc ua ing demands, ensu ing mo e
e icien and esponsi e sys em pe o mance.
Table 5. Reques La ency and Tail Pe o mance
Reques Ra e ( eq/sec)
La ency (P50)
La ency (P90)
La ency (P99)
Blue-G een
190ms
310ms
470ms
Cana y
140ms
210ms
260ms
Cana y deploymen s consis en ly
main ained lowe la ency ac oss all
pe cen iles, pa icula ly unde bu s and
sus ained loads, demons a ing i s abili y o
handle a ic demands e icien ly. In con as ,
Blue-G een deploymen s expe ienced
signi ican ly highe la ency, wi h P99 la ency
nea ly 80% highe , highligh ing poo ail-end
pe o mance du ing load shi s. This dispa i y
unde sco es he bene i s o Cana y's
inc emen al ollou app oach, which helps
isola e aul s and s abilize la ency. By
g adually in oducing changes, Cana y
deploymen s minimize he isk o
o e whelming he clus e , ensu ing mo e
consis en and eliable pe o mance. In
con as , Blue-G een's all-o -no hing app oach
can lead o pe o mance deg ada ion du ing
a ic swi ches.
Table 6. Rollback E iciency
Me ic
Blue-G een
Cana y
Rollback Time
6.5 seconds
9.2 seconds
Failed Reques s Du ing Rollback
18% spike
<5% spike
Blue-G een deploymen s enabled
nea -ins an ollbacks by simply swi ching
a ic ou es, o e ing a speedy eco e y
op ion. Howe e , Cana y deploymen s,
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
50
al hough sligh ly slowe in ollback,
demons a ed a signi ican ad an age in
minimizing ailed eques s. This was achie ed
h ough eal- ime me ic- igge ed ollbacks
applied o a pa ial subse o a ic, ensu ing a
mo e con olled and sa e e e sal p ocess.
The insigh he e is ha while Blue-G een
ollbacks a e as e , hey can be iskie wi hou
p ope moni o ing. In con as , Cana y
ollbacks, hough sligh ly slowe , a e
inhe en ly sa e and mo e aul - ole an ,
making hem pa icula ly sui able o eal- ime
sys ems whe e eliabili y and p ecision a e
c ucial.
Table 7. Th oughpu and Load Handling
Scena io
Blue-G een ( eq/sec)
Cana y ( eq/sec)
Sus ained Load
140 eq/sec
190 eq/sec
Bu s Load Handling
Delayed
S able a 170+
Cana y deploymen demons a ed
supe io sus ained h oughpu , pa icula ly
when in eg a ed wi h KEDA (e en -based
au oscale ) and Is io o ou ing. This
combina ion allowed o e icien handling o
a ic demands. In con as , Blue-G een
deploymen s uggled o s abilize a e ab up
a ic shi s, esul ing in b ie pe iods o
h oughpu deg ada ion. The key insigh is ha
Cana y deploymen suppo s high- eloci y
con inuous deli e y and handles e en -d i en
loads mo e g ace ully. By g adually
in oducing changes and scaling esou ces
acco dingly, Cana y ensu es a mo e s able and
e icien sys em pe o mance, making i well-
sui ed o en i onmen s wi h dynamic a ic
pa e ns.
The esul s conclusi ely show ha
Cana y deploymen s, despi e hei mo e
complex se up, ou pe o m Blue-G een
deploymen s in Kube ne es en i onmen s in
e ms o scalabili y, esilience, and ollback
con ol. While Blue-G een deploymen s a e
well-sui ed o speci ic use cases such as low-
equency ull sys em upda es o scena ios
equi ing ins an ull ollback, Cana y
deploymen s a e mo e adap able o mode n,
dynamic en i onmen s. Cana y's s eng hs
make i pa icula ly sui able o eal- ime
mic ose ices, equen code deli e y, and
da a-d i en Si e Reliabili y Enginee ing (SRE)
en i onmen s, whe e obse abili y and
p og essi e ollou a e essen ial. By le e aging
Cana y's capabili ies, eams can achie e mo e
eliable and e icien deploymen s, be e
aligning wi h he demands o con empo a y
so wa e de elopmen and ope a ions.
Limi a ions and Fu u e Resea ch:
This s udy, despi e i s comp ehensi e
na u e, is subjec o se e al limi a ions ha
wa an acknowledgmen . Fi s ly, he
expe imen s we e conduc ed in a con olled
Kube ne es clus e wi h simula ed a ic,
which may no ully eplica e he complexi ies
o eal-wo ld p oduc ion en i onmen s whe e
ac o s like ne wo k la ency and in e -se ice
dependencies play a ole. Addi ionally, he
s udy's ocus on s a eless mic ose ices limi s
i s gene alizabili y o s a e ul se ices o
applica ions wi h di e en ansac ional
dynamics. The eliance on speci ic ools like
Is io and P ome heus may also in oduce bias,
as al e na i e con igu a ions could yield
di e en ou comes. Fu he mo e, he sho -
e m obse a ion window may o e look long-
e m pe o mance implica ions such as
memo y leaks o au oscale ecalib a ion.
IJAAR Vol. 6 No. 38 ISSN – 2347-7075
Ami K. Mogal & Vaibha P. Sonaje
51
Finally, he single-clus e se up does no
accoun o he complexi ies o mul i- egion o
hyb id cloud deploymen s, whe e ne wo k
la ency and ou ing could impac deploymen
s a egy e ec i eness. These limi a ions
highligh a eas o u u e esea ch o u he
alida e and ex end he indings.
Fu u e esea ch di ec ions building
upon his s udy's indings could explo e
se e al key a eas. Fi s ly, e alua ing Blue-
G een and Cana y deploymen s in mul i-
clus e and edge en i onmen s would p o ide
insigh s in o hei pe o mance unde
geog aphically dis ibu ed scena ios.
Secondly, in eg a ing AI/ML-d i en
au oscale s could op imize ollou and ollback
p ocesses. Thi dly, in es iga ing he impac o
hese deploymen pa e ns on s a e ul and
s eaming applica ions would be aluable.
Addi ionally, examining secu i y and
compliance conside a ions, de eloping cos
op imiza ion models, and explo ing human-in-
he-loop ollback s a egies could u he
enhance he unde s anding and applica ion o
hese deploymen s a egies in eal-wo ld
con ex s
Conclusion:
The e olu ion o mic ose ices and
cloud-na i e a chi ec u es has made
deploymen s a egies a c i ical componen o
sys em eliabili y, pe o mance, and agili y.
This esea ch unde ook a comp ehensi e
compa a i e s udy o wo widely adop ed
Kube ne es deploymen models—Blue-G een
and Cana y—wi h a speci ic ocus on hei
scalabili y, ollback e iciency, and ope a ional
esilience. Th ough a se ies o con olled
expe imen s in ol ing simula ed wo kloads,
a ic spikes, and induced ailu e scena ios,
he s udy e ealed ha Cana y deploymen s
o e supe io scalabili y and aul isola ion,
pa icula ly in dynamic and high-concu ency
en i onmen s. Cana y's g adual ollou
mechanism, when pai ed wi h eal- ime
eleme y and au oscaling logic, allowed o
mo e adap i e sys em beha io and lowe ail-
end la ency. Al hough ollbacks in Cana y
deploymen s we e sligh ly slowe han in
Blue-G een, hey we e mo e a ge ed and
esul ed in ewe se ice dis up ions. In
con as , Blue-G een deploymen s excelled in
en i onmen s whe e simplici y, p edic abili y,
and ull ollback speed we e mo e c i ical han
esou ce e iciency. Thei ease o se up and
bina y a ic swi ching model made hem
well-sui ed o monoli hic o low- equency
elease pipelines. Howe e , hey incu ed
highe in as uc u e o e head due o
en i onmen duplica ion and showed
limi a ions unde bu s and sus ained a ic
loads. Ul ima ely, he s udy demons a es ha
nei he s a egy is uni e sally supe io , bu
a he , he op imal choice depends on he
deploymen con ex , isk ole ance, and
obse abili y ma u i y o he o ganiza ion. Fo
eams p io i izing ine-g ained con ol,
p og essi e deli e y, and minimal blas adius,
Cana y deploymen is he p e e ed model. Fo
o ganiza ions equi ing apid ull-sys em
swi ches wi h limi ed in as uc u e
complexi y, Blue-G een emains a iable
op ion. This wo k con ibu es o he ongoing
discou se on De Ops and cloud-na i e
deploymen bes p ac ices by p o iding da a-
d i en insigh s, eal-wo ld me ics, and
empi ical compa isons be ween deploymen
s a egies. I also lays a ounda ion o u u e
explo a ion in o mul i-clus e , AI-enhanced
au oscaling, and hyb id deploymen models
ha can u he op imize he balance be ween
speed, sa e y, and e iciency in mode n
applica ion deli e y pipelines.
