We scaled down some post transaction batch jobs (status sync consumers) which freed up a chunk of DB connections to normal, reduced contention for the tables. The dependency of legacy EC service was turned off to allow EC to prevent it from holding up further connections due to upstream latencies.

We have initiated restart of all application pods across major services which were affected the most (txn, customer, wallets and Saved Payment methods). We are continuously monitoring the key metrics.

We are seeing an increase in 504s across all the production k8s clusters.