Memory and CPU spike on 3.13.3 gerrit version

[Gagantej Rao]

Hi Gerrit community, 

 Recently we tried upgrading to 3.13.3 from 3.12.1 post which we observed drastic  Memory and CPU spike during the regular load. Below the environment details 

We run gerrit as docker container with HA setup active - active ( for clustering we use PCS + DLM ). Also we use AWS 

Instance size: m7i.2xlarge ( 3 instances ) 
volume: ebs and type io2 ( shared across 3 instances ) 
Clustering:  PCS + DLM 

gerrit config is below 

[gerrit]
    basePath = git
    serverId = "1233-5667-789"
    canonicalWebUrl = "https://git.uici.com/"
    instanceId = "git"
[container]
    user = gerrit
    javaHome = /opt/java/openjdk
    heapLimit = 24g
    javaOptions = "-Dflogger.backend_factory=com.google.common.flogger.backend.log4j.Log4jBackendFactory#getInstance"
    javaOptions = "-Dflogger.logging_context=com.google.gerrit.server.logging.LoggingContext#getInstance"
    javaOptions = -Djava.security.egd=file:/dev/./urandom
    javaOptions = --add-opens java.base/java.net=ALL-UNNAMED
    javaOptions = --add-opens java.base/java.lang.invoke=ALL-UNNAMED
[auth]
    type = OAUTH
    gitBasicAuthPolicy = HTTP
    logoutUrl = https://sso.com/idp/startSLO.ping?LogoutType=AsyncOnly
    cookiedomain = ".uic.com"
[receive]
    enableSignedPush = false
    timeout = 6min
[sendemail]
    from = "\"Gerrit\" <noreply@.com>"
    smtpServer = server
    smtpServerPort = 587
    smtpEncryption = tls
    smtpUser = "username"
    html = false
[sshd]
    threads = 32
    listenAddress = *:29418
    maxConnectionsPerUser = 0
    backend = NIO2
[httpd]
    listenUrl = proxy-https://*:8080
    maxQueued = 1000
    minThreads = 10
    maxThreads = 50
[cache]
    directory = cache
[gitweb]
    cgi = /usr/share/gitweb/gitweb.cgi
    type = gitweb
[change]
    submitWholeTopic = true
    mergeabilityComputationBehavior = API_REF_UPDATED_AND_CHANGE_REINDEX
[core]
    packedGitOpenFiles = 4096
[gc]
    startTime = Sat 23:59
    interval = 1 week
[lfs]
    plugin = lfs
    backend = fs
[index]
    type = lucene
[accountPatchReviewDb]
    url = "jdbc:postgresql://db:5432/gerritdatabase?user=gerrit&password=gerrit1234"
[receive]
    autogc = false
[plugins]
    allowRemoteAdmin = true
[plugin "emoticons"]
    showEmoticons = true
[plugin "code-owners"]
    disabled = true
    requiredApproval = Code-Review+1
    overrideApproval = Owners-Override+1
    allowedEmailDomain = ui.com
    allowedEmailDomain = test.com

we have tried reducing heap, restricting threads and increasing resources to m7i.4xlarge but didnt help. Any suggestion on this to fix the issue.

During the load we may have 25 pipelines cloning and sending review from CI pipelines. around 50 - 60 dev's can clone and push their changes. 

Any help on this is much appreciated.

Thanks 
Gagantej