Helm Virtual index resolution will frequently hang in HA deployments when concurrent requests get routed to different nodes

Reproduce:

1.

a=1; while [ $a -le 10 ]; do curl "http://<LOAD_BALANCER_URL>/artifactory/helm-virtual/index.yaml" > $a & a=`expr $a + 1`; done

2. List the creation time of the downloaded files, notice the times are quite staggered, meaning it took too long to serve some of those requests.

ls -l -T

Running this several times will cause some of the requests to hang up to 2 minutes (the Work Queue timeout to wait for an async invocation), or until another request to the same index.yaml (same underlying WorkQueue unique key) goes to the node where the hung requests were routed which "revives" the queue, but only if its actually able to acquire a lock on that unique key.