[#21963] YSQL: New read restart workloads.

[pao214]

New Workloads

Workload	Short Description
SqlEventCounter	Track counters for events. Eg. clicks, views, etc.
SqlBankTransfers	Transfer money between accounts and verify balances checkout.
SqlWarehouseStock	New Orders decrement stock and a concurrent process restocks low stock items.
SqlMessageQueue	Stock levels is disseminated by the publisher and the subscriber accumulates the total stock.
SqlConsistentHashing	Find the node closest to key using consistent hashing in a dynamic node configuration.

---

Benchmark: Restart Read Requests

Workload	Wall Clock	Clockbound Clock
SqlEventCounter	~8000	~110
SqlBankTransfers	~500	~3
SqlWarehouseStock	~150	~8
SqlMessageQueue	~50	~5
SqlConsistentHashing	~1500	~10

---

Security

Change the prometheus URL to http since we start yugabyted without TLS by default. Security is not the primary concern of this change.

---

Cluster Setup

1. Create 3 AWS nodes on US N Viriginia with permissive networking.
2. Copy/Download release tarball with LTO enabled onto the machines.
3. Extract the tarball and cd into the directory.

For wall clock configuration,

./bin/yugabyted start [--join <first_node_private_ip>]

For clockbound clock configuration,

./bin/yugabyted configure clockbound
./bin/yugabyted start --enhance_time_sync_via_clockbound

Run each sample app for 3 times for benchmarking and confidence in the metrics.

[pilshchikov]

Maybe in doRead workload do a read and check consistency? And this logic can be moved in doWrite method and activate on a coin flip?

[pao214]

I check if there are any stale reads now. The name doRead is unfortunate but it is better to report latency and throughput separately for different operations. If we use a coin flip, then the throughput, latency numbers mix across different operations.