Player is loading...
Managing an Akka Cluster on Kubernetes
Markus Jura, MOIA
Wednesday, June 12, 2019 · 3:33 p.m. · 38m 23s
Akka Cluster is a powerful toolkit for building distributed applications. But as always, power comes with a price. Special care needs to be taken during bootstrapping, network partitions and gracefully stopping a cluster. In this talk, I'll introduce Akka Management and Akka Split Brain Resolver modules and show how they can help to manage an Akka Cluster in production. Join us when you want to know how a stateful application can handle rolling deployments without downtime and without losing in-flight requests and how it can safely recover from network partitions.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
uh_huh
00:00:05
welcome everyone to join and if you're not yet yeah i'm very proud to be
00:00:10
here the tends anniversary of scott days it's also actually the tense anniversary of arco
00:00:17
um got i think the first version room was released in two thousand nine um and they also gave
00:00:23
a talk first time it's gotta days so therefore i thought it's a great idea to talk about our cup
00:00:29
and i would talk about how to manage in our cluster on one company does
00:00:36
i'm not sure i'm i'm i'm doing start and i can offer eight yes um
00:00:40
started um at the at the start up of mine and um then i have
00:00:46
worked for enough years for for like bent mainly on the conductor team which is
00:00:51
also an orchestration to a a similar to company does or two that was him marathon
00:00:57
and since two years i'm working for more yeah which is right chang service
00:01:01
and we as not only loving scallop but it's it's basically something like what
00:01:07
but where where people share the same right and uh therefore we reduce the congestion
00:01:13
of cities and it's also for the electric so we really try to be green
00:01:17
and the main programming language at at more yeah it's it's got up
00:01:22
and uh we uh we already have launched in hamburg as well as
00:01:25
i know in hanover in germany and uh more cities will will come
00:01:33
so uh before we jump into the talk i just want to have a disclaimer
00:01:37
because it's a talk about distributed systems and i don't want to ruin your life if
00:01:43
if you never really worked deeply was distributed system so now it's a chance to do so room um
00:01:49
because just dos you uh uh does you one time set and i think that we need to
00:01:54
it can really in distributed systems is post accelerating and frustrating
00:02:00
when things work it's like a symphony when they don't it's like a elevens birthday party we have skits on speed
00:02:08
and i would you really think it's true if you want to um dude
00:02:14
in that debugging um it's it's really hard so you really need to make the things right
00:02:21
and that's also partially what my talks about um especially if you deploy an art class on company does you need to
00:02:27
consider three things and i will um yeah guide you through
00:02:31
through them one is how to how to form an ah cluster
00:02:36
the second thing is how to handle a rolling updates uh how to
00:02:41
um perform according nato graceful shut down and also how to hannah network petitions
00:02:48
so let's start with bootstrapping um the orchid team created a very good library um i think to use
00:02:55
from now it's called r. can management and also consist of another lie required adjustable
00:03:00
strap and um the nice thing about that is that doesn't use or doesn't need any
00:03:07
central coordination searches to people it c. d. before we had libraries such as
00:03:12
constructor uh by my by my uh um yeah friend hike was a beggar
00:03:17
um conduct or set some solutions uh but you always needed a sentry coordination system such as to keep our
00:03:24
and are just a bootstrapping and doesn't need that and uh
00:03:29
i will explain basically how it works and what doesn't need it
00:03:33
so that two faces and the first phase is the discovery phase so
00:03:37
we have here on on the left and the right uh two notes
00:03:42
and um basically so first thing they do is say they want to get to know
00:03:46
each other right what is your i. p. address please give me your i. p. address
00:03:50
so for the node one and no two they somehow need to get the i. p. addresses from each other how they're doing that
00:03:57
the different discovery seconds you can use was adjustable strap one is
00:04:01
uh the quinn you just eighty i another one is the n. s. for
00:04:04
instance if younger we need as i would recommend the company does eighty i
00:04:09
and the receiving the i. p. addresses from each other um by yeah
00:04:13
going to the company does eighty i based on the ports so this name
00:04:18
if you acted system equal it's basically the parts of his name then
00:04:22
you're good to go then access to put strip figure that out automatically
00:04:29
so they basically then no ooh there eighty addresses the contact points and once they know that
00:04:37
no cluster has been forget right they're not joining it just a at the only know each other
00:04:42
so then the second phase comes in its cord contact probing
00:04:47
and basically each note will will ask each other please give me a seat notes right the seat
00:04:53
node is basically was an i. p. address when a class the already has been formed uh then
00:04:59
you should we just as the i. p. address as a seat node and then
00:05:03
i don't know what's can join the existing cluster so are a a class the
00:05:09
puts rip basically exposes an h. t. t. endpoint puts proceed notes where you
00:05:14
can get the currents seat notes that a certain market cluster node is aware of
00:05:20
and then there to use cases one use cases the no cluster has been forget
00:05:27
so then like a node would say yes then no seat notes um
00:05:32
and obviously there's also use case joining an existing cluster so we would
00:05:36
go through both of them so it's an all seed notes available yet
00:05:40
then the second step is that a certain state a timeout and we'll pass um
00:05:49
that means that no other contact point uh should be added or should be removed
00:05:54
i'm also no seat node should be added and one says
00:05:57
stay the time margin has passed in new cluster would be formed
00:06:03
and someone needs to form that this cluster right and market
00:06:06
cluster basically one note it's safe on its needs to join itself
00:06:12
how's artist a bootstrapping doing that it just taking the lowest i. p. address in the previous face
00:06:19
in gathering the contact point the i. p. addresses um
00:06:22
every node knows now like the i. p. addresses off my
00:06:26
of my part of the different instances so they also know every node knows which is the lowest i. p. address
00:06:33
so then basically the note with the lowest i. p. address we just form a cluster with itself
00:06:39
and by doing so it will also we just itself as a seat not
00:06:44
and then the other notes they will keep continuing um yeah
00:06:48
executing the h. t. t. p. c. notes and point and then
00:06:53
the second node will then uh we get back that there's the
00:06:58
signal one and therefore it will also join the existing cluster so that's
00:07:03
technique we oppose use cases are covered without using any
00:07:08
coordination service or just to keep our white c. d.
00:07:12
and yeah the academe implemented that and uh you just need to use it i
00:07:18
will show you in a demo how you configure your your application our demo set up
00:07:22
is the following we have um not application um we have it when you just keep going and
00:07:28
configuration so that orchid application is running in inside of parts and then we also the company to service
00:07:36
uh which is basically which basically groups the different arcana
00:07:40
cluster instances in kind of a micro service and it also is responsible for deciding
00:07:48
to what kind of instance the h. e. p. traffic is routed
00:07:53
um and then on top of that in our demo we also have in in ingress which is basically an eighty
00:07:59
i gateway to expose also our h. e. d. p. server
00:08:03
to to the outside so that we can interact with it
00:08:08
so wish me luck with the dental x.
00:08:13
just let it sink
00:08:16
huh
00:08:20
but but not really doing so much should be fine okay
00:08:27
so first off that's um that's just see it in
00:08:31
action and then we see at the actual configuration um
00:08:39
or no actually we we do with the way around so first i show you the company does um configuration
00:08:45
so in order to to read a lark application inside of a part we need to define
00:08:51
you couldn't you just configuration and all the company does consecration is done in younger files so this easy on the fire
00:08:57
keep going and younger and we need to we need to define a deployment resource and
00:09:03
basically one important thing is that you need to specify here and the at name
00:09:08
as as metadata and this should be the same name as
00:09:12
as you act assistance so that our cluster bootstrap um yeah we'll
00:09:17
we'll figure that out and we use some other configuration we come
00:09:22
to that in a minute um but also what you need to define
00:09:27
our is a docket image so basically our our application to stalk arrest with s. p. t. native package up
00:09:35
and i'm using your local don't doctor registry so i'm i just need to specify in my company
00:09:40
does conflagration the the docket much um and then we also need to specify ports are three ports so
00:09:52
obviously our h. t. p. so well um we need to set up it's on port eighty eighty um
00:09:59
then the icon management port that is necessary that says
00:10:03
seat not h. t. t. p. endpoints available throughout other parts
00:10:09
so that they can figure out the signalled information and also
00:10:12
the arco remote import we got because we have our cluster so
00:10:16
and are classed as communicating by our remote so you also need to open the sport
00:10:21
this conflagration doesn't mean that it's exposed to the outside
00:10:25
it just visible inside of you are inside of reports
00:10:32
alright and the rest of the configuration we we see we check out later then in the application can take
00:10:40
for the first discovery phase you need to specify kind of the discovery back and
00:10:44
uh we have chosen the company does a. p. i. to figure out the the contact points and then
00:10:53
also you need to specify some puts the configuration here we have um
00:10:59
defined required contact point number of three we have a three note cluster and
00:11:04
basically we only go to the second phase of of forming a cluster or
00:11:09
joining an existing cluster if at least three contact points are visible by each node
00:11:15
right so all of them and then i've also talked about just even margin so we wait five seconds um
00:11:22
before we going into into the into the probing face um where
00:11:27
no other contact point should be added or or signature in it i
00:11:33
alright and then in our um coat it's very simple we just need
00:11:39
to execute two lines of code the first one is to study aachen management
00:11:45
and the second one um is yeah starting the orchestra puts for procedure
00:11:51
right with soccer management you get um you also get
00:11:55
well additional information was an h. e. p. a. p. i. way you can
00:12:00
rick rest um the current archive members and and so forth
00:12:04
and a bootstrap really triggers the start of your of your application
00:12:12
so now let's run it um in order to run it into any just you
00:12:19
you just apply the the the younger file right on the planet you may find it and what
00:12:27
i will also do here in the second line is watching the parts of our we need to cluster
00:12:35
right now we don't have any ports deployed and now if i run the amplifier uh we spin up three notes
00:12:43
and we also for instance can have a look at the lock
00:12:49
and as you can see here they're also not ready
00:12:52
yet so they're starting up now and once basically um
00:12:57
yeah we can serve h. e. p. traffic they will become ready so now they already took twenty two seconds
00:13:05
so now let's see if we can access it i'm i'm using h. t.
00:13:10
p. i. n. to um to make a trip from design to hamburg and
00:13:17
it doesn't work because we haven't configured the company disservice
00:13:22
in ingress yet right the as i said these ports
00:13:27
of the parts um the exposed on the part letter but not um to
00:13:31
the outside world and he was mini cue by cue we're basically getting the um
00:13:36
basically yeah the outside cool b. i. p. of the minute you cluster so what we also need
00:13:42
to do is um configuring two additional company does
00:13:46
resources so those and and ingress i've group them in
00:13:52
in another on the fire code all going and they're very very easy so
00:13:59
basically here the first one is the service uh you give it a name
00:14:03
you give it the selector that's important that needs to be a part
00:14:05
name and um then you also need to map support right our ports
00:14:11
port of our h. c. p. server was eighty eighty and we're mapping it now to to port eighty
00:14:16
um and then we also need to define ingress resource um where we basically
00:14:23
say okay please use the trip service that we just defined here um and expose it to port eighty
00:14:31
and we also applied in similar fashion just using should control acute cut to um apply
00:14:40
and then let's let's see if that has been started yes and the service
00:14:51
yes
00:14:54
so now the h. t. t. p. should go through yeah and we have greater than we're trip for ten years from wasn't ten bucks
00:15:04
an hour more pixels up north and fourteen not you need to
00:15:08
go to and look to experience it okay cool um so that's oops
00:15:16
huh that's put scripting
00:15:22
so the second part is um yeah coordinated shut down so it's really really important ah cluster
00:15:29
that you shut down your application in the right order um that you shut down
00:15:35
adjust the shopping are just to sing it and also your h. t. t. p.
00:15:39
and maybe your incoming i'm keen uses or kafka events um in in the right order
00:15:49
so usually that is pretty easy right you just need to ensure
00:15:54
um if you're not using a cluster that you drain you h. t. t. p. connections and
00:15:59
you are as a crown as you when it's connection such as backing is s. q. s. kafka
00:16:05
and then afterward you also need to ensure that you process the in flight h. e. p.
00:16:10
request or um request that have been sent over
00:16:14
over a yeah yawn as as encounters events um
00:16:20
but when you deal with an attack just it gets a bit more complicated
00:16:24
we have here on the left hand side that's a note eight and on the right hand side note b.
00:16:29
and what can happen is that basically a cheeky and racing events
00:16:34
has been already um yeah drained but then in you message comes in
00:16:39
from note p. over our clustering let's say using our cluster starting
00:16:44
and your shot entity your use uh uh we m. is living here on the
00:16:48
left hand side so then basically this um yeah our remote will send a message to
00:16:54
two node a even though no day isn't in termination more to write um
00:17:00
we see that here so you also need to take here that your training
00:17:04
basically this channels is incoming channel and luckily also the okay team
00:17:09
uh worked on on a small module record our car coordinated shut down that you
00:17:14
should use if you're using our cluster that takes care of that at least if
00:17:20
in your article just a shot entity or in your pocket cluster singing
00:17:24
actor ewan not calling another external service over future for instance which you
00:17:31
anyways shouldn't do inside of an actor um if you are having such scenario where inside of you are
00:17:38
attack last um charted actor for instance you're calling other
00:17:42
things other services then you should rather use marcus streams and
00:17:47
hike we actually gave a talk um today morning uh august screens to the extreme uh where he also use that
00:17:54
within we are using carcass screams i'm using sing crafts um to
00:18:00
two hundred that properly but if you if you basically stay
00:18:03
local inside of your cluster charting act or sing it nectar
00:18:07
uh then you'll find with this talk and i will explain you how that works
00:18:11
if you if you going again remote inside of it then watch the top i call
00:18:17
okay so the article denied that shut on module is inside of the um
00:18:22
yeah act or actor quality and it consists of different shut down faces
00:18:30
um their cup of shut down faces registered for classes starting
00:18:34
forcing it and that you that you should not use for you
00:18:38
as the application developer um i would say and number two and number three of the most important ones service and
00:18:44
bind to stop except i am to stop accepting incoming request
00:18:49
and so was request done to to process in flight requests
00:18:55
and you should basically we're just a custom shop download it both
00:19:01
for h. t. p. as well as for your as in chronos events
00:19:06
we're just bring 'em is pretty easy so you just
00:19:10
basically take take the coordinated shut down it's an extension
00:19:14
um take take the handler and then just call at
00:19:17
task and then you need to define a certain face um
00:19:22
like face before class started on give it a name and then in york in the closer at your custom stop logic
00:19:30
also sometimes you want to trigger a coordinated shut down so that would be kind of the replacement of system not exit
00:19:38
right you should new system dot x. it because it exits the j. v.
00:19:41
m. directly instead you want to follow the order in the coordinated shut down so
00:19:45
um then you should use the run method to trigger the actual coordinated shut down and as you as we have seen
00:19:53
sorry as we have seen here the last step is actually terminating the actor system and the j. d. m.
00:19:59
so don't you system that exit and also don't use um
00:20:04
yeah it j. v. m. o. shut down hole instead use here the task to register your custom stop logic
00:20:12
that's about it um the here's an example with dark h. t. t. p. so
00:20:18
you you're using the run method to directly trigger the shut down in case the
00:20:24
and the address and port has been already pound when we're starting in h. c. p. server and um if
00:20:31
basically the h. c. p. server has been started successfully then what sorry then we're registering a task in the face
00:20:40
face so this week rest done that's right face for for processing in flight requests
00:20:46
and also um the ark a team okay g. p. team has
00:20:51
has it just implemented to terminate method that you can use it is taking care of
00:20:58
not allowing any new connections anymore over the h. e. p. sockets and then
00:21:02
afterwards it will it will basically process the in flight request that has to
00:21:08
on going on your application and once or the in flight requests have been processed
00:21:13
it will it will basically a complete with the with the future of them um if
00:21:20
it hasn't been completed and maybe like some some in flight requests taking
00:21:24
very long then you also need to specify a if in iteration um
00:21:30
and then at some point in time it will basically yeah a terminate
00:21:35
in the hard way and then you might use this particular in in fact requests
00:21:40
so also take yeah that is termination deadline um yeah is is large enough
00:21:48
so that was only the ark apart of coordinated shut down together with
00:21:52
company does um they also some tricks so let's also go through them
00:21:57
so let's say we're doing a rolling update and then an existing parts
00:22:03
will receive a shot on signal
00:22:07
then what you can do is you can register it three stop folk in quinn eaters and
00:22:13
this priest topple folk you can execute some code and only after this code has been
00:22:19
executed it with four to the the sick turn signal the shot on signal to the port
00:22:26
and you can use that in order to
00:22:30
um ensure that the readiness check failed so when he does have some
00:22:34
his checks we will also um look at that in a minute and
00:22:39
the readiness check is important only if if the part is ready um the company to service
00:22:45
we're route traffic to the sport and if the part is not ready anymore
00:22:50
uh it will not become any or we not get any traffic anymore so
00:22:53
he and the shut down basically we ensure first at the redness check as fate
00:22:58
so that no h. e. p. traffic is routed to this instance anymore um and then on the h.
00:23:04
t. on the r. k. g. p. side we only need to ensure to process the in flight requests
00:23:11
with a ten minute method um and then if you
00:23:14
use our clustering you automatically use the coordinated shut down
00:23:20
extension law clustering is using it so then uh a car yeah basically the
00:23:25
the market cluster would be shut down in the right order afterwards the cluster
00:23:30
uh would be left and then at the end the actor system in j.
00:23:34
b. m. would be to i mean it it so let's see that an action
00:23:48
so i missed most of my
00:23:52
my tuning yeah but no problem i so i've already created a
00:24:01
new version of our application and in order to deploy the new version
00:24:08
we are um yeah we're just executing again the q. control applied it
00:24:20
deployments configuration but before we're doing that i first in
00:24:27
in the to clone and younger fire need to change the version
00:24:32
so that's what we need is configuration and so in
00:24:36
order to see something i will also watch the parts again
00:24:42
and we also would check the dogfight or the existing um
00:24:48
yeah parts that will now be terminated and should yeah i
00:24:55
and the nice thing about we need is is basically when i
00:24:58
when i use is cheap uh control apply function it will check what
00:25:03
kind of deployment is currently running and what is the content of the
00:25:07
new department uh we have changed version so it will automatically figure out
00:25:12
um that it needs to perform the rolling department because that's also configured in
00:25:17
in the deployment amplifier which we check in detail in a minute
00:25:21
survey execute that then what you can see here is um
00:25:26
that we spin out a new instance and three instances
00:25:30
as to running and asked to ready and i'm only if
00:25:37
basically you the force node is is ready then we
00:25:40
start terminating an existing note um for small clusters we have
00:25:46
just here three note just uh uh that's a good strategy where you do they will go from three to four
00:25:52
from four to three three to four four two three and only um yeah basically upgrade
00:25:58
one node how after the other as you can see here um yeah the shut down is
00:26:06
works so in total it should take around want a minute if you have a larger cluster it also fine too
00:26:14
to spin up multiple instances in power uh um so if your final
00:26:19
seven nine fifty fifty two node cluster
00:26:25
okay now we just perform the rolling update
00:26:29
so let's look uh what you need to do to each each that so first
00:26:34
let's have a look at the company does configuration um we need to define the strategy
00:26:40
rolling update and was isn't mac search one and max and availability
00:26:45
that's basically the different definition that you just some makes and availability means that
00:26:50
we always should have three instances running that are serving h. e. p. traffic
00:26:56
and make search means that we only spin up one new node
00:27:01
at the time so that's a very conservative definition but it works for small clusters
00:27:06
as in our example and then let's also have a look at our pretty stop hawk
00:27:12
um but also very simple um so we first create
00:27:17
a shot on fire on the file system to kind of
00:27:21
signal to the ark application that the shut down is now in progress and then we're waiting
00:27:26
not phone arbitrary amount of time um but for a time where
00:27:32
the readiness check as fate and also some some part time so
00:27:37
um another confer gauge configuration you can see here is that we have to find a readiness has check
00:27:46
i'm a custom h. t. p. endpoint and basically the
00:27:50
readiness fates once three times as h. t. p. endpoint
00:27:54
has given not it two hundred successful um and it
00:28:00
will also try in into of all of two seconds so
00:28:03
three multiplied by two is six and then we also give it like
00:28:07
two seconds more time that we really ensure here that is pretty stop
00:28:11
hawk is only finished once no connections um yeah um
00:28:17
i basically wanted to this part that we want to terminate
00:28:24
and then in our application cons um we we need to configure
00:28:33
coordinated chart on saying that that's the j. v. m. should exit
00:28:38
at the end so you should do that when you own company does and you also should give it
00:28:43
uh before it's time out for your faces that you would just about yourself like on our k. g. p. so
00:28:51
if we look in on all okay g. p. so that it's basically similar what you've just seen on the slides
00:29:01
and um yeah basically research registering here are shut down task in the face
00:29:08
so this request done and um insuring that the in flight request
00:29:14
i fully processed
00:29:23
okay that last one is handling network partitions
00:29:28
so what you really should avoid is a is a split brain scenarios when you're dealing with an how cluster
00:29:36
and how does it generally work with with an target just if you have a network petition so for instance if you own a double yes
00:29:44
and you and the same a obvious region but are using different availability
00:29:48
zones you already in different data centres um so network partitions can occur and
00:29:55
he in our case we have network petition um where d. c. one and d. c. three states each other
00:30:02
uh but they don't see d. c. to anymore and market clustering is using our remote thing to always
00:30:09
check under the what if um if every noticed a
00:30:12
reachable right and if the network petition is ongoing obviously
00:30:17
uh d. c. one and d. c. three there we're not c. d. c.
00:30:20
to anymore so what what should happen then like should dizzy to still be alive
00:30:28
so the right choices that that we should down one of the one of the sites
00:30:33
either the left side or the right seven usually you down um the minority side of notes
00:30:40
because otherwise what could happen is if i don't know you
00:30:44
using are just the shopping and you're trading the same user um
00:30:50
one one d. c. too and also on the c. one and then laid on the network petition goes away you haven't in
00:30:56
costa inconsistent state you have created two times the same user
00:31:00
you have created to chart entities are with in our justice writing
00:31:04
uh so you always need to ensure that that you down the notes and in fact
00:31:08
our car i like a cluster i'm always try to market these nodes and as unreachable
00:31:16
to mark notes as a reachable um marcus using inferior detector
00:31:21
and basically it will use a gossip record to parody kiosk hey
00:31:26
a dizzy to are you still there um and in this case the
00:31:30
c. one would say no i'm i cannot see d. c. to anymore
00:31:34
and then it would uh sends information to d. c. three and d. c. three would also maybe check out if
00:31:42
if uh they can does can c. d. c. two and so on so at the end of a a detector
00:31:47
takes care of that um well that d. c. one
00:31:51
and this is three marks deceit within reachable and vice versa
00:31:55
and our car is using the egg true if a a detector for that if you would
00:31:59
like to more about that um minor dad also give a pretty good talk last year um about
00:32:06
about more detail it's of the fayette detector
00:32:11
and this fe detector we mark the notes as i'm reachable it doesn't mean that they're downed yet
00:32:16
and that they uh go on it only means that
00:32:20
that they cannot see each other and that is kind of a stalemate state in our cluster
00:32:26
when notes on reachable you cannot know new joint you know
00:32:30
new uh nodes can join um and also are just uh
00:32:38
charlene cannot we balance the notes from one note to another for instance so you need
00:32:43
to take care of that as i reach this state is only for short period of time
00:32:48
and uh yeah usually what you should do is either shut down it's
00:32:53
manually um was shut down like a side um programmatically
00:32:59
and to shut down notes programmatically um yeah ah chi has basically an interface
00:33:04
uh which you can implement or libraries can implement and
00:33:08
the scored split brain resolution or split brain was oliver
00:33:13
and the different libraries out there and you also can use
00:33:17
different strategies so one strategy if you have a cluster that's
00:33:22
uh that is rather static so always five node always a seven node cluster
00:33:26
then you should use static wrong where you define um basically number of of majority
00:33:33
notes um let's say we have our three not just uh if i define too
00:33:37
then always the nodes that's each other uh they will still remain
00:33:42
and the minus side will be downed if you have more flexible
00:33:48
um scaling scenario where you quite often scaling up from five to
00:33:52
nine to thirteen down to seven then you should use icky majority
00:33:57
strategy which um yep was always down to the minority side
00:34:02
and there's also keep oldest innkeeper free if you have a light and subscription i
00:34:07
would recommend to to use a light and split brain result of a library um it's
00:34:12
it's obviously always maintained it's it's better test that it works with all the different strategies
00:34:18
but they also some open source libraries out there if you don't have a subscription they don't work
00:34:23
for strategies um but all of them are working for the uh for the static for at least
00:34:29
um if you want to know more about that i'm just yeah come
00:34:34
to me afterward um we uh we are using our custom custom downing
00:34:42
the source code is on on the top um if you want to follow me
00:34:47
one i that would all get up um it says handle and um we also if you if you're looking for a job
00:34:55
obviously we also hiring and we also have a breezy it's colour day so just
00:34:59
come to me or come to all those uh later on thank you very much
00:35:11
and i think we have to ten minutes more time for questions
00:35:19
uh_huh
00:35:22
i i know that many people use shouting we
00:35:25
wheeze presses comes from the purses come form on cassandra
00:35:31
ah emerges or else could be rain scenarios how do you
00:35:35
deal with the fact that customer is making different kind of credence
00:35:39
the sons that uh in my understanding the the remain
00:35:42
a available um partition current are a drop the consistency
00:35:50
yeah so i mean also when you use cassandra you always need to choose you are you're consistent strategy right if you
00:35:57
doing the read and during the ride oh if you want to have like a
00:36:01
ride or where you ensure that that was the nodes needs to be there and consistent
00:36:06
um if you have a network petition this would fail but you
00:36:09
also can just ride on a on a chrome level for instance um
00:36:13
and then the ride would succeed on and the same with the read so you always need to do and shoes basically
00:36:20
a consistency that if you want a strong consistency um that we're not work in uh when you have a network petition
00:36:28
and yeah so you you're very flexible with cassandra in that case
00:36:43
yeah hi when you and perform the routing updates all start class the
00:36:48
how do you deal with the c. we're see we're isolation all done
00:36:53
the messages it's a new cluster nodes and all class the notes
00:36:58
yeah that's a good question so and it's also in
00:37:00
the arcade ox if if you have a breaking changes in
00:37:07
basically phones in your case classes that you sending over the wire why our remote right
00:37:11
marcus assigning for instance um then you first need to down the entire class that um
00:37:17
so and then spin of the new classes either use a blue green you put on it
00:37:21
um if you want to have no downtime um why you're shutting down your entire cluster and
00:37:28
then you also need to apply virgin burton strategy um of your
00:37:33
case laws so for instance you might see realise that was probably both
00:37:37
and and then you have probably you might be deployed instead where you first introduced a new version
00:37:44
like aversion to um which is then basically deployed and
00:37:49
available on all the new notes um and
00:37:54
and then later on um basically also removing the whole week note
00:37:58
uh sorry the old version in your in your portable schema for instance
00:38:08
anyone
00:38:12
okay then thanks very much enjoy discoveries
Conference Program
A Tour of Scala 3
Martin Odersky, Professor EPFL, Co-founder Lightbend
June 11, 2019 · 5:15 p.m.
8337 views
A story of unification: from Apache Spark to MLflow
Reynold Xin, Databricks
June 12, 2019 · 9:15 a.m.
1267 views
In Types We Trust
Bill Venners, Artima, Inc
June 12, 2019 · 10:15 a.m.
1569 views
Creating Native iOS and Android Apps in Scala without tears
Zahari Dichev, Bullet.io
June 12, 2019 · 10:16 a.m.
2231 views
Techniques for Teaching Scala
Noel Welsh, Inner Product and Underscore
June 12, 2019 · 10:17 a.m.
1295 views
Future-proofing Scala: the TASTY intermediate representation
Guillaume Martres, student at EPFL
June 12, 2019 · 10:18 a.m.
1156 views
Metals: rich code editing for Scala in VS Code, Vim, Emacs and beyond
Ólafur Páll Geirsson, Scala Center
June 12, 2019 · 11:15 a.m.
4695 views
Akka Streams to the Extreme
Heiko Seeberger, independent consultant
June 12, 2019 · 11:16 a.m.
1552 views
Scala First: Lessons from 3 student generations
Bjorn Regnell, Lund Univ., Sweden.
June 12, 2019 · 11:17 a.m.
577 views
Cellular Automata: How to become an artist with a few lines
Maciej Gorywoda, Wire, Berlin
June 12, 2019 · 11:18 a.m.
386 views
Why Netflix ❤'s Scala for Machine Learning
Jeremy Smith & Aish, Netflix
June 12, 2019 · 12:15 p.m.
5026 views
Massively Parallel Distributed Scala Compilation... And You!
Stu Hood, Twitter
June 12, 2019 · 12:16 p.m.
958 views
Polymorphism in Scala
Petra Bierleutgeb
June 12, 2019 · 12:17 p.m.
1113 views
sbt core concepts
Eugene Yokota, Scala Team at Lightbend
June 12, 2019 · 12:18 p.m.
1655 views
Double your performance: Scala's missing optimizing compiler
Li Haoyi, author Ammonite, Mill, FastParse, uPickle, and many more.
June 12, 2019 · 2:30 p.m.
837 views
Making Our Future Better
Viktor Klang, Lightbend
June 12, 2019 · 2:31 p.m.
1682 views
Testing in the postapocalyptic future
Daniel Westheide, INNOQ
June 12, 2019 · 2:32 p.m.
498 views
Context Buddy: the tool that knows your code better than you
Krzysztof Romanowski, sphere.it conference
June 12, 2019 · 2:33 p.m.
393 views
The Shape(less) of Type Class Derivation in Scala 3
Miles Sabin, Underscore Consulting
June 12, 2019 · 3:30 p.m.
2321 views
Refactor all the things!
Daniela Sfregola, organizer of the London Scala User Group meetup
June 12, 2019 · 3:31 p.m.
514 views
Integrating Developer Experiences - Build Server Protocol
Justin Kaeser, IntelliJ Scala
June 12, 2019 · 3:32 p.m.
551 views
Managing an Akka Cluster on Kubernetes
Markus Jura, MOIA
June 12, 2019 · 3:33 p.m.
735 views
Serverless Scala - Functions as SuperDuperMicroServices
Josh Suereth, Donna Malayeri & James Ward, Author of Scala In Depth; Google ; Google
June 12, 2019 · 4:45 p.m.
936 views
How are we going to migrate to Scala 3.0, aka Dotty?
Lukas Rytz, Lightbend
June 12, 2019 · 4:46 p.m.
709 views
Concurrent programming in 2019: Akka, Monix or ZIO?
Adam Warski, co-founders of SoftwareMill
June 12, 2019 · 4:47 p.m.
1974 views
ScalaJS and Typescript: an unlikely romance
Jeremy Hughes, Lightbend
June 12, 2019 · 4:48 p.m.
1377 views
Pure Functional Database Programming‚ without JDBC
Rob Norris
June 12, 2019 · 5:45 p.m.
6374 views
Why you need to be reviewing open source code
Gris Cuevas Zambrano & Holden Karau, Google Cloud;
June 12, 2019 · 5:46 p.m.
484 views
Develop seamless web services with Mu
Oli Makhasoeva, 47 Degrees
June 12, 2019 · 5:47 p.m.
785 views
Implementing the Scala 2.13 collections
Stefan Zeiger, Lightbend
June 12, 2019 · 5:48 p.m.
810 views
Introduction to day 2
June 13, 2019 · 9:10 a.m.
250 views
Sustaining open source digital infrastructure
Bogdan Vasilescu, Assistant Professor at Carnegie Mellon University's School of Computer Science, USA
June 13, 2019 · 9:16 a.m.
374 views
Building a Better Scala Community
Kelley Robinson, Developer Evangelist at Twilio
June 13, 2019 · 10:15 a.m.
245 views
Run Scala Faster with GraalVM on any Platform
Vojin Jovanovic, Oracle
June 13, 2019 · 10:16 a.m.
1340 views
ScalaClean - full program static analysis at scale
Rory Graves
June 13, 2019 · 10:17 a.m.
463 views
Flare & Lantern: Accelerators for Spark and Deep Learning
Tiark Rompf, Assistant Professor at Purdue University
June 13, 2019 · 10:18 a.m.
380 views
Metaprogramming in Dotty
Nicolas Stucki, Ph.D. student at LAMP
June 13, 2019 · 11:15 a.m.
1250 views
Fast, Simple Concurrency with Scala Native
Richard Whaling, data engineer based in Chicago
June 13, 2019 · 11:16 a.m.
624 views
Pick your number type with Spire
Denis Rosset, postdoctoral researcher at Perimeter Institute
June 13, 2019 · 11:17 a.m.
245 views
Scala.js and WebAssembly, a tale of the dangers of the sea
Sébastien Doeraene, Executive director of the Scala Center
June 13, 2019 · 11:18 a.m.
661 views
Performance tuning Twitter services with Graal and ML
Chris Thalinger, Twitter
June 13, 2019 · 12:15 p.m.
2003 views
Supporting the Scala Ecosystem: Stories from the Line
Justin Pihony, Lightbend
June 13, 2019 · 12:16 p.m.
163 views
Compiling to preserve our privacy
Manohar Jonnalagedda and Jakob Odersky, Inpher
June 13, 2019 · 12:17 p.m.
301 views
Building Scala with Bazel
Natan Silnitsky, wix.com
June 13, 2019 · 12:18 p.m.
565 views
Diversity and Inclusion Panel: Bring the Thunder!
Panel
June 13, 2019 · 2:30 p.m.
244 views
Asynchronous streams in direct style with and without macros
Philipp Haller, KTH Royal Institute of Technology in Stockholm
June 13, 2019 · 3:45 p.m.
304 views
Interactive Computing with Jupyter and Almond
Sören Brunk, USU Software AG
June 13, 2019 · 3:46 p.m.
681 views
Scala best practices I wish someone'd told me about
Nicolas Rinaudo, CTO of Besedo
June 13, 2019 · 3:47 p.m.
2702 views
High performance Privacy By Design using Matryoshka & Spark
Wiem Zine El Abidine and Olivier Girardot, Scala Backend Developer at MOIA / co-founder of Lateral Thoughts
June 13, 2019 · 3:48 p.m.
753 views
Immutable Sequential Maps – Keeping order while hashed
Odd Möller
June 13, 2019 · 4:45 p.m.
276 views
All the fancy things flexible dependency management can do
Alexandre Archambault, engineer at the Scala Center
June 13, 2019 · 4:46 p.m.
389 views
ScalaWebTest - integration testing made easy
Dani Rey, Unic AG
June 13, 2019 · 4:47 p.m.
468 views
Mellite: An Integrated Development Environment for Sound
Hanns Holger Rutz, Institute of Electronic Music and Acoustics (IEM), Graz
June 13, 2019 · 4:48 p.m.
213 views
Closing panel
Panel
June 13, 2019 · 5:54 p.m.
400 views
Recommended talks
PERFMATTERS for Android
Hasan Hosgel, ImmobilienScout24 / Berlin, Germany
Nov. 27, 2016 · 11:45 a.m.
The Young Software Engineer’s Guide to Using Formal Methods
K. Rustan M. Leino, Amazon
June 7, 2018 · 10:16 a.m.
963 views