Application Engineering

Question 1

What are the components of a database server?

Answer 1

• CPU
• Memory - inserts and updates are written here first
• Persistent Disk
• Journal - log of every database process

Database writes to memory first and then writes to journal and disk. There is a lag between the write to memory and the write to disk.

Question 2

How could the data be lossed when updating and inserting even though Mongo returns a message saying insert/update success?

Answer 2

The server can crash after the data is written to memory but before it is written to disk. It’s up to the developers to configure Mongo for waiting for a persistance to disk before receiving a successful save message.

Question 3

What is the default write concern for MongoDB?

Answer 3

w=1 j = false –> Save occurs for a write to the memory in the database and there’s no wait for the journal to sync

Question 4

How do you set a wait for the write to disk in Mongo?

Answer 4

w = 1 j = true –> This waits for the write to disk when saving which is much slower than w=1 j = false option, but less risky.

Question 5

Provided you assume that the disk is persistent, what are the w and j settings required to guarantee that an insert or update has been written all the way to disk?

Answer 5

w=1, j=1

Question 6

What typically happens when you don’t receive a affirmative response?

Answer 6

Network error; this can also give a false positive

Question 7

What are the reasons why an application may receive an error back even if the write was successful?

Answer 7

Network connection between the server and application and the server was reset after the write was received but before the message was sent
Mongodb can also terminate between receiving the response and writing it to disk
Mongodb can fail between the time of the write and the time the application receives a response.

Inserts instead of updates is the only way to prevent this

Question 8

How do we get availability and fault tolerance in MongoDB?

Answer 8

Using replication across multiple Mongo instances. Data written to a primary node will asynchronously be replicated to the secondary nodes

Question 9

What is the minimum original number of nodes needed to assure the election of a new Primary if a node goes down?

Answer 9

3 –> You need two other secondaries when a primary goes down for an election. When the primary goes down, the a new primary is elected and the data is written there. The old primary returns as a secondary.

Question 10

What are the different types of replica set nodes?

Answer 10

1. Regular - has data; can be primary or secondary
2. Arbiter - only there for voting purposes in elections
3. Delayed/Regular - disaster recovery node (can’t participate in voting - p = 0)
4. Hidden - can’t be a primary but can participate in elections

Question 11

Which types of nodes can participate in elections of a new primary?

Answer 11

Regular
Hidden
Arbitors

Question 12

Explain Write Consistency

Answer 12

• Writes go to the only primary but reads don’t have to go to the primary
• ## If reads go some where else, you may read stale data

Question 13

During the time when failover is occurring, can writes successfully complete?

Answer 13

No

Question 14

What is the oplog?

Answer 14

It is a capped collection on a mongo instance (you can find it using ‘show collections’) that follows the activity of the primary. The activity of the primary is recorded on the primary’s oplog and then these are copied by the secondaries asynchronously

• doesn’t matter which driver you use with these - version doesn’t matter or driver doesn’t matter (WiredTiger)

Question 15

What does rs.slaveOk() do?

Answer 15

It allows a secondary instance to be queried on; by default they can’t

Question 16

T or F –> A copy of the oplog is kept on both the primary and secondary servers

Answer 16

T

Question 17

T or F –> Replication supports mixed-mode storage engines For example mmpap1 and wiredTiger

Answer 17

T

Question 18

What happens if a node comes back up as a secondary after a period of being offline and the oplog has looped on the primary?

Answer 18

The entire dataset will be copied from the primary

Question 19

T or F - There is a scenario that a write performed with w = majorithy get rolled back

Answer 19

T - because this can happen when a failover on the primary occurs after the write is committed to the primary but before the replication of the oplog is committed to the secondaries. This will show an exception on the client.

Question 20

If you leave a replica set node out of the seedlist within the driver, what will happen?

Answer 20

It will be discovered anyways as long as you list atleast one valid node

Question 21

What kind of exception occurs when a primary fails and election occurs during an insert?

Answer 21

AutoReconnect exception

Question 22

What will happen if the following statement is executed in Python during a primary election?

Answer 22

db.test.insert_one({‘x’:1})

Question 23

How is failover detected?

Answer 23

Use catches and exceptions in your code to catch a failover - this however does not guarentee that your failover catch will write all data - instead use a retry up to 3 times or more

Question 24

Why don’t you need to handle a duplicate key exception for reads?

Answer 24

Duplicate key reads are impossible when reading

Question 25

Give examples of item potent updates and non item potent updates:

Answer 25

Itempotent –> $set
NonItempotent –> $inc, $push –> these are more risky to run again because they could potentially doubly increment - not good for incrementing sensitive data like money but probably not a big deal for

Question 26

How do you handle not item potent updates?

Answer 26

• Converting to item potent updates
• ## Not updating again and not caring about it

Question 27

If you want to be sure that an update with a $inc occurred exactly once in the face of failover, what’s the best way to do it?

Answer 27

transform the update into a statement that is item potent - however this risks losing one update in a multithreaded program