How to optimize reading of many versions of delta table

[processadd]

Hi,
I have a raw delta table raw loaded from Kinesis by a structured streaming job, and every several seconds there is a new version created in raw, appended from the structured streaming job.
I also have a structured streaming job with trigger=availableNow to read from raw and it is triggered daily. So each trigger might see thousands of versions from raw. I use foreachBatch to do MERGE on a target delta table.
The MERGE seems fast (less than 1 min) but it shows a long time on duraionMs like

[batchId = 320] Streaming query made progress: {
  "id" : "...",
  "runId" : "...",
  "name" : null,
  "timestamp" : "2024-10-21T02:52:57.029Z",
  "batchId" : 320,
  "batchDuration" : 182297,
  "numInputRows" : 0,
  "inputRowsPerSecond" : 0.0,
  "processedRowsPerSecond" : 0.0,
  "durationMs" : {
    "addBatch" : 51186,
    "commitOffsets" : 66,
    "getBatch" : 37802,
    "latestOffset" : 91964,
    "queryPlanning" : 783,
    "triggerExecution" : 182287,
    "walCommit" : 317
  },
  "stateOperators" : [ ],
  "sources" : [ {
    "description" : "DeltaSource[s3://...]",
    "startOffset" : {
      "sourceVersion" : 1,
      "reservoirId" : "...",
      "reservoirVersion" : 3292332,
      "index" : -1,
      "isStartingVersion" : false
    },
    "endOffset" : {
      "sourceVersion" : 1,
      "reservoirId" : "...",
      "reservoirVersion" : 3294059,
      "index" : -1,
      "isStartingVersion" : false
    },
    "latestOffset" : null,
    "numInputRows" : 0,
    "inputRowsPerSecond" : 0.0,
    "processedRowsPerSecond" : 0.0,
    "metrics" : {
      "numBytesOutstanding" : "705134",
      "numFilesOutstanding" : "28"
    }
  } ],
  "sink" : {
    "description" : "ForeachBatchSink",
    "numOutputRows" : -1
  }
}

Why it needs so long time for latestOffset and addBatch?

Thanks