1
00:00:01,230 --> 00:00:04,860
Come back still with relational database service or are they.

2
00:00:04,890 --> 00:00:13,200
And let's dive into a replica as and how you can maintain more or scale on the heat capacity of the

3
00:00:13,200 --> 00:00:14,020
database.

4
00:00:14,040 --> 00:00:19,710
So when the required read all capacity is reached because of intense limitation because of the number

5
00:00:19,710 --> 00:00:21,730
of applications that are carrying that.

6
00:00:21,780 --> 00:00:23,790
But you still need more capacity.

7
00:00:23,790 --> 00:00:27,690
So what do you do then use or take advantage of the replicates.

8
00:00:27,750 --> 00:00:29,040
So what is that replica.

9
00:00:29,190 --> 00:00:33,170
We have as we mentioned before we have the primary and we have this standby right.

10
00:00:33,300 --> 00:00:37,710
Or it could be only the problem and if it is a stand alone implementation so that either case can be

11
00:00:37,710 --> 00:00:45,070
created from a stand alone against instance or from the primary of a multi easy or D s deployment.

12
00:00:45,090 --> 00:00:50,940
So there is basically whatever is written to the primary we know before that it will be synchronously

13
00:00:50,940 --> 00:00:57,840
replicated to the standby but it will be asynchronously replicated then to replicate.

14
00:00:58,040 --> 00:00:59,320
So there is some delay here.

15
00:00:59,340 --> 00:01:06,090
They call it time lag between when the information is written to the prime rate until it becomes available

16
00:01:06,090 --> 00:01:11,640
on the reader because it's asynchronous replication not synchronous or prickish for this primary to

17
00:01:11,640 --> 00:01:15,760
become the source of replication to replicates.

18
00:01:15,870 --> 00:01:18,830
It must have automatic backups enabled.

19
00:01:18,840 --> 00:01:25,420
So basically it must have a retention period of more than zero any value more than zero.

20
00:01:25,560 --> 00:01:32,300
That means automated backups are running kernel you have read replicas made of a multi RDX deployment.

21
00:01:32,310 --> 00:01:36,810
Yes but they would be made from the prime minister the primary would be the source and it will then

22
00:01:36,810 --> 00:01:41,940
create or replicate to that he is going to have multiple region because yes you can have multiple reader

23
00:01:41,940 --> 00:01:47,990
because for all the read replicas you can have up to 15 Orrery because that can be distributed in the

24
00:01:47,990 --> 00:01:55,320
availability zones where the cluster is but within a region so it cannot span beyond the region that

25
00:01:55,420 --> 00:02:00,240
the Republicans will be within the use cases for a BECAUSE WHY DO I NEED THEM.

26
00:02:00,270 --> 00:02:02,960
If you want to shift to intensive application.

27
00:02:02,970 --> 00:02:07,890
So instead of reading that it intends to application see the freedom from the primary.

28
00:02:07,890 --> 00:02:14,010
Now I will shift the applications to read from exactly from one or more triplicates if I want to do

29
00:02:14,010 --> 00:02:20,280
data warehousing then shifted or analytics then shifted to the real replicas and make all that operations

30
00:02:20,280 --> 00:02:25,020
happen from here instead of overloading that primary database instance.

31
00:02:25,170 --> 00:02:31,930
Use Case number two is killing beyond the capacity of your main database instance for heavy workload

32
00:02:32,220 --> 00:02:37,470
and service the traffic while the source is unavailable let's say for any reason you have the primary

33
00:02:37,470 --> 00:02:43,710
and the primary is either under attack it's it's going through a support process maintenance or anything

34
00:02:43,710 --> 00:02:46,110
like that or it cannot scale it.

35
00:02:46,140 --> 00:02:48,340
It has reached its limit from the.

36
00:02:48,390 --> 00:02:50,180
What would that be Dipika help me with.

37
00:02:50,220 --> 00:02:57,090
So now it can read from the replicas because they will still be available even if the source iOS on

38
00:02:57,220 --> 00:03:04,160
Saturday or they will just kill the requirement because you cannot have more than four instances involving

39
00:03:04,160 --> 00:03:05,340
that application chain.

40
00:03:05,700 --> 00:03:12,150
So as you mentioned my sequel and maybe have is they support the replicas of real replicas.

41
00:03:12,150 --> 00:03:14,530
How many can I can I really could.

42
00:03:14,550 --> 00:03:15,660
So I have here.

43
00:03:15,810 --> 00:03:20,980
The source has replicated so this is level one of each application.

44
00:03:21,000 --> 00:03:22,620
Can I get to a second level.

45
00:03:22,620 --> 00:03:29,730
Yes you can connect it to a third level yes you can connect it to a fourth level no income up more than

46
00:03:29,730 --> 00:03:33,240
four levels can not be in a chain of prickish.

47
00:03:33,240 --> 00:03:38,460
Now remember one thing you can do this but remember that here is asynchronous.

48
00:03:38,460 --> 00:03:44,310
And then when the data gets here of data then this is asynchronous and when it gets updated then this

49
00:03:44,310 --> 00:03:51,150
is asynchronous so you have legs and legs and Laxey need to be mindful of the potential delay.

50
00:03:51,180 --> 00:03:56,640
The data will be available on that in that region because if you change them like that a potential exam

51
00:03:56,640 --> 00:04:01,220
question let's say we have a multi is the primary standby.

52
00:04:01,250 --> 00:04:03,220
And I have created three replicas.

53
00:04:03,420 --> 00:04:07,210
OK so I have here multiple one or more Republicans.

54
00:04:07,440 --> 00:04:10,050
Now all of them are replicating from the source right.

55
00:04:10,080 --> 00:04:11,520
They are reading from the source.

56
00:04:12,870 --> 00:04:18,240
What happens if there is a fail over manual or automatic from the primary to a standby.

57
00:04:18,240 --> 00:04:22,490
So this one becomes the primary and this one which is the standby.

58
00:04:22,830 --> 00:04:24,930
Do they still get their application from here.

59
00:04:24,930 --> 00:04:26,110
The answer is No.

60
00:04:26,310 --> 00:04:31,210
They will be shifting to the other one as the application source.

61
00:04:31,230 --> 00:04:36,240
So if the first instance of a multi deployment fails over the secondary to the secondary any associated

62
00:04:36,710 --> 00:04:43,140
with because our switch to use the secondary as the replication source with replicas are they automatically

63
00:04:43,290 --> 00:04:46,100
deleted when I delete a database instance.

64
00:04:46,110 --> 00:04:49,840
If I go ahead and delete the primary order is it because I'm going to go away.

65
00:04:49,840 --> 00:04:52,120
No you have to do it manually as well.

66
00:04:52,320 --> 00:04:55,040
What happens if I delete the primary.

67
00:04:55,170 --> 00:04:56,580
So I have the primary source.

68
00:04:56,580 --> 00:04:59,630
I get the source but I forgot to delete it.

69
00:04:59,650 --> 00:05:01,330
Replicates what will happen now.

70
00:05:01,380 --> 00:05:08,300
They will automatically all get promoted to exactly a single easy stand alone database instance you'll

71
00:05:08,300 --> 00:05:15,780
end up having one two three for instance in different availability zones if you promote my sequel or

72
00:05:15,780 --> 00:05:21,060
my database region because that is in turn if we get into another replica those three because we remain

73
00:05:21,090 --> 00:05:21,840
active.

74
00:05:21,840 --> 00:05:29,610
So here we have a primary and I have in my mind equal or my database so I have read Level 1 and this

75
00:05:29,610 --> 00:05:33,070
was replicating to replicate level 2.

76
00:05:33,660 --> 00:05:39,530
Now if I promote what you mean by promote If I change this one to become a fully fledged are the first

77
00:05:39,570 --> 00:05:43,500
instance to read and write what happens to this one it doesn't die.

78
00:05:43,530 --> 00:05:45,480
Does it go from the primary.

79
00:05:45,480 --> 00:05:46,410
Does it go away.

80
00:05:46,410 --> 00:05:49,090
What happens if you promote this one.

81
00:05:49,290 --> 00:05:52,030
This one will continue to lead to this.

82
00:05:52,070 --> 00:05:57,990
Anyways I was the source of applications that promote this this will continue as is and it will continue

83
00:05:57,990 --> 00:06:05,900
to read from the new standalone ideas in cross region read replicas and let's say I have a standby multi

84
00:06:05,890 --> 00:06:08,440
easy in one region or one.

85
00:06:08,520 --> 00:06:17,130
Now if I'm very concerned about disaster recovery can I create in region are two replica of our primary

86
00:06:17,130 --> 00:06:21,060
database instance that exists in region one a different region.

87
00:06:21,090 --> 00:06:27,900
If you have a region and you have multiple ACS and you would like to build a fault tolerant solution

88
00:06:28,020 --> 00:06:31,570
that would even cater for a regional failure.

89
00:06:31,620 --> 00:06:39,630
So think of immediately for them are the US think of a cross region or read replicas and this is available

90
00:06:39,630 --> 00:06:41,290
for my mom post.

91
00:06:41,550 --> 00:06:48,120
My sequel maybe and possibly a sequel so it can be used to provide for lower RPO.

92
00:06:48,270 --> 00:06:54,420
So basically when you do that three replicas if you lose a region then you can have since you are doing

93
00:06:54,420 --> 00:06:58,260
replication into triplicated it will take you a few minutes to bring this up and running.

94
00:06:58,260 --> 00:07:03,450
So that means your data loss will be less and that's why you were saying lower RPO.

95
00:07:03,450 --> 00:07:07,280
Another way this could be asked about in a in a in a question.

96
00:07:07,290 --> 00:07:15,270
Let's say you have an enterprise that has presence in many regions but in the HQ region in the main

97
00:07:15,570 --> 00:07:20,830
company our headquarter region they would like to run sales analytics themselves reporting.

98
00:07:21,000 --> 00:07:27,180
So you can query and have an application here that needs to query all these different databases.

99
00:07:27,180 --> 00:07:32,430
They are are the result in different regions I mean and you want to query all of this and that will

100
00:07:32,430 --> 00:07:34,020
be query intensive as well.

101
00:07:34,320 --> 00:07:39,350
And then you can under-reporting and then the reports will be ready for the management.

102
00:07:39,360 --> 00:07:44,040
Why don't you as a solution to avoid latencies over the Internet and all that.

103
00:07:44,040 --> 00:07:51,540
Why don't you just create a replica of each one in the HQ region and then run your application on an

104
00:07:51,700 --> 00:07:57,110
X across or against that reads replicas that are available within the same region.

105
00:07:57,210 --> 00:07:58,500
That will give you two benefits.

106
00:07:58,500 --> 00:08:03,000
The first one is you don't have to burden the artist that the business with the queries.

107
00:08:03,030 --> 00:08:09,540
The second one is you avoid latencies and they available here to run that and you provide reporting

108
00:08:09,690 --> 00:08:10,340
quicker.

109
00:08:10,410 --> 00:08:13,790
You can promote read pick up promoting Republicans to stand alone.

110
00:08:13,890 --> 00:08:19,800
So if you have read the replica in one easy or in one region that you would like for some reason to

111
00:08:19,800 --> 00:08:25,550
promote this to become a stand alone single everliving it is on every instance.

112
00:08:25,560 --> 00:08:31,080
So I'm not going to be a multi easy one then you can do that you can select that and then promoted to

113
00:08:31,080 --> 00:08:33,970
become standalone fully fledged database.

114
00:08:34,110 --> 00:08:35,250
Now what I do.

115
00:08:35,260 --> 00:08:39,970
What do I do if I hit the largest or this instance and I still need to skate.

116
00:08:40,020 --> 00:08:41,130
What do I do then.

117
00:08:41,130 --> 00:08:48,040
There is a way to do partitioning and to split your data base over multiple odious instances.

118
00:08:48,060 --> 00:08:50,820
However this goes back to the limitation we talked about.

119
00:08:50,820 --> 00:08:56,580
Remember we talked about the non-relational and we said that it horizontally skills perfectly.

120
00:08:56,580 --> 00:08:59,690
This is a limitation of our ideas and then we start going through workarounds.

121
00:08:59,700 --> 00:09:05,850
If you want higher faster and on that day there is in probably if you crossed the all the limitations

122
00:09:05,850 --> 00:09:12,240
of all the limits of the US probably you may need to consider that an addition that is if possible about

123
00:09:12,300 --> 00:09:18,940
cloud it trace logging or calls to the API of the order of service and of course the cloud trail will

124
00:09:18,940 --> 00:09:23,380
log that for your API calls to the odiously Debby's instance.

125
00:09:23,380 --> 00:09:29,180
Let's not talk about the ideal usage patterns or the patterns and type patterns for our ideas.

126
00:09:29,200 --> 00:09:35,470
Are This is the idea for existing applications that rely on my sequel Oracle or sequel server a traditional

127
00:09:35,470 --> 00:09:41,110
relational database engines and one good reason for that is that our ideas offers full compatibility

128
00:09:41,110 --> 00:09:43,890
and direct access to native database engines.

129
00:09:43,900 --> 00:09:49,840
Most code libraries and tools designed for these databases and treat work and modified with Amazon are

130
00:09:49,840 --> 00:09:50,220
the best.

131
00:09:50,230 --> 00:09:56,590
So basically when you bring a sequel from on premise or Oracle or sequel as long as audience doesn't

132
00:09:56,590 --> 00:10:01,690
have any limitations for the features you are trying to use then you can bring on your database.

133
00:10:01,720 --> 00:10:06,930
If it is my sequel sequel server or Microsoft or Oracle into our ideas and you should be fine.

134
00:10:06,960 --> 00:10:13,570
Amazon audience is also optimal for new applications with structured data that requires more sophisticated

135
00:10:13,570 --> 00:10:15,170
queening enjoining capability.

136
00:10:15,280 --> 00:10:22,420
Anti-pattern this is very important when not to use it when to avoid Ardis if an audience does not provide

137
00:10:22,480 --> 00:10:24,950
a database platform you're looking for.

138
00:10:25,090 --> 00:10:31,560
So give me an example IBM D-B to Informix Sybase or any other database that are not supported by our

139
00:10:31,560 --> 00:10:35,950
ideas at the moment when you want to deploy the database or religion to be in the cloud.

140
00:10:35,980 --> 00:10:39,610
Then you will have to go for a self-managed database on MSNBC too.

141
00:10:39,640 --> 00:10:45,370
So the alternative if it is not supported by our s or like it Oracle RAC when you have some limitations

142
00:10:45,460 --> 00:10:50,740
and these are not supported by the US then you need to deploy your oracle rack or similar databases

143
00:10:50,770 --> 00:10:53,160
or IBM or Informix or Sybase.

144
00:10:53,170 --> 00:10:57,800
You need to deploy it on easy to instances and you manage it and control it yourself.

145
00:10:57,800 --> 00:11:00,070
So how can you do that how can you implement it yourself.

146
00:11:00,070 --> 00:11:01,720
You need a relational database.

147
00:11:01,720 --> 00:11:08,970
A.m.-I so from IBM or Informix so I've been able to or by installing software on an Amazon easy to instance

148
00:11:08,980 --> 00:11:10,290
so that is the first one.

149
00:11:10,290 --> 00:11:15,640
Complete control if you would like to have complete control including level control of the database

150
00:11:15,640 --> 00:11:21,970
server with full route or admin logging privileges then you need to deploy your letters on easy to what

151
00:11:21,970 --> 00:11:28,030
else what other limitations do we have or type patterns of index and query forecast data.

152
00:11:28,030 --> 00:11:34,900
So if it's only about indexing and query focused so complex sophisticated querying or complicated joining

153
00:11:34,900 --> 00:11:40,270
of tables then many cloud based solutions will not require or may not require especially that we'd be

154
00:11:40,320 --> 00:11:45,760
the ones that those features in relation database such as joint and complex transaction so if you don't

155
00:11:45,760 --> 00:11:48,640
need that then definitely dynamic to me would be a better option.

156
00:11:48,640 --> 00:11:52,780
In your case numerous blocks What is the blob's binary large objects.

157
00:11:52,780 --> 00:11:57,040
And as you said these are like videos images audio files.

158
00:11:57,360 --> 00:12:02,390
One of the engines provides an audio support binary large objects blobs.

159
00:12:02,410 --> 00:12:09,430
If your application makes heavy use of them so you have a solution that is all about users uploading

160
00:12:09,430 --> 00:12:11,470
images and uploading videos and all that.

161
00:12:11,470 --> 00:12:13,820
So in that case don't go for ideas.

162
00:12:14,080 --> 00:12:18,930
You can't have indexing or pointers in a database.

163
00:12:18,970 --> 00:12:26,800
And if that was intended because then it's even preferable to be on TV than are the US and save your

164
00:12:27,340 --> 00:12:30,000
images and files and all the blogs on SD.

165
00:12:30,010 --> 00:12:36,130
So basically you index or you query the database for a pointer and then you fix that pointer and go

166
00:12:36,130 --> 00:12:38,350
ahead and get it from S 3.

167
00:12:38,560 --> 00:12:39,840
Automated scalability.

168
00:12:39,910 --> 00:12:45,040
If you want automated scalability and fully automated one and then there will be a better one for you.

169
00:12:45,040 --> 00:12:47,080
So that was about the US.

170
00:12:47,110 --> 00:12:48,720
So let's take a break again.

171
00:12:48,730 --> 00:12:49,290
Thank you.