$8.$ In Bigtable, how does the Primary$/$Master discover the current tablet assignment when it starts?

Data discovery and assignment of tablets

Each client uses the BigTable library. There is one master and then there are many tablet servers. Each tablet server is assigned ownership of certain tablets by the master. Master is also doing load balancing and then assigning unassigned tablets to available tablet servers. Lets dive into that a bit.

When operations on some rows$($tablets$)$ is needed, master checks to see if those tablets are being served by some tablet$-$server. Master looks this up in two$-$level index$($METADATA tables$)$ which contains information of tablet assignments. This location information is handled out the clients and then clients can directly communicate with tablet servers and thus eliminating load from the master. This two level hierarchy allows for pretty large table sizes of upto $2$ MB $(^2*^2*200$MB$)($You can refer to the paper for further details on to arrive at this number$)$

Discovery of tablets

Master assigns a tablet to a tablet server. When a tablet server has capacity, master can send the $$load tablet$$ to the tablet server. The master knows about functioning and available tablet server because when a tablet server comes up$,$ it takes an exclusive lock on chubby file. When the tablet server dies, it releases the lock on the chubby file. Master is obviously monitoring these chubby directory namespace and files. So when a tablet server dies, the master can reassign the tablet to another tablet server. Paraphrase this answer