Cache storage policy application results (Search, Replicate)

[cthulhu-rider]

code and unit tests are ready, intergration tests pass. I also want to take a memory profile in devenv

---

i created 5 containers and ran 30 routines sending 1000 search queries (each time to random container)

v0.42.1

current branch

---

Go script

func TestDevenv(t *testing.T) {
	strCnr := []string{
		"3SDKVLon6bk6ALm8PCEjaGHUeg8cP5d6cvSJdpuMJ3qy",
		"5gGHUYnAXfB7y5XSFJb4SBr2g3NvEnkAELDAV9rxFdWL",
		"65NX4euCiJDDkduNeSc8puo7oKa3ddZC2dHxNMSS3UBA",
		"9EPJGmDDDv9iNqP1rAy7pyYuaJHRy2XSyMZt81DL35cD",
		"ArUd3NXPUhaDLNZuEKYHjbc4GS7u4nViDTKbyLiDtENW",
	}
	const endpoint = "s01.neofs.devenv:8080"
	const workers = 30
	ctx := context.Background()
	signer := test.RandomSignerRFC6979(t)

	cnrs := make([]cid.ID, len(strCnr))
	for i := range strCnr {
		require.NoError(t, cnrs[i].DecodeString(strCnr[i]))
	}

	c, err := client.New(client.PrmInit{})
	require.NoError(t, err)
	var prm client.PrmDial
	prm.SetServerURI(endpoint)
	require.NoError(t, c.Dial(prm))
	t.Cleanup(func() { c.Close() })

	var wg sync.WaitGroup
	st := time.Now()
	for i := 0; i < workers; i++ {
		wg.Add(1)
		go func() {
			defer wg.Done()

			for i := 0; i < 1000; i++ {
				if i%100 == 0 {
					fmt.Println(i)
				}
				rdr, err := c.ObjectSearchInit(ctx, cnrs[rand.Int()%len(cnrs)], signer, client.PrmObjectSearch{})
				if err != nil {
					log.Println("init search:", err)
					rdr.Close()
					continue
				}
				err = rdr.Iterate(func(id oid.ID) bool { return false })
				if err != nil {
					log.Println("read search response:", err)
					rdr.Close()
					continue
				}
				rdr.Close()
			}
		}()
	}
	wg.Wait()
	fmt.Println(time.Since(st))
}

Potential

besides obvisuly required implementing the same approach to other components/RPC which is gonna TBD in next PRs, i got following thoughts:

1. the past data is less in demand than the current one, so less space can be allocated for it
2. alternatives to LRU strategy
3. dynamic volume estimation taking into account the number of requested containers and storage nodes in the network
4. storing indices instead of node info structures (netmap: Provide placement methods returning indices instead of copied node descriptors neofs-sdk-go#541)
5. invalidate cached results on container removal. Normally, they will be automatically evicted according to the LRU principle. But it’s still better to force the displacement
6. in proposal, the cache is invalidated by changing the epoch. However, the network map may not change, or the change does not affect the policies of some containers

[codecov]

Codecov Report

Attention: Patch coverage is 96.00000% with 2 lines in your changes missing coverage. Please review.

Project coverage is 23.73%. Comparing base (4f90cc2) to head (10d05a4).
Report is 3 commits behind head on master.

Files	Patch %	Lines
cmd/neofs-node/policy.go	96.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2892      +/-   ##
==========================================
+ Coverage   23.67%   23.73%   +0.05%     
==========================================
  Files         775      775              
  Lines       44908    44933      +25     
==========================================
+ Hits        10633    10664      +31     
+ Misses      33420    33417       -3     
+ Partials      855      852       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[carpawell]

Also, the last commit says it was tried to be done unsuccessfully before, can you, please, provide more info about it (to the commit)?

[cthulhu-rider]

Also, the last commit says it was tried to be done unsuccessfully before, can you, please, provide more info about it (to the commit)?

done