I’v read that spark session context is thread-safe but not in all cases.

I have multi thread application which is organized so :

N – workers serving event-bus and sending some simple Spark tasks .

Now i want to add job description to each task to make pipeline more observable using something like :

  def withJobDesciption[T](desc:String)(fn: => T)(implicit spark: SparkSession) : T = {
    spark.sparkContext.setJobDescription(desc)
    try {
      fn
    } finally {
      spark.sparkContext.setJobDescription("")
    }
  }

sparkContext looks like global storage and must not be shared between threads. From the other point of view i see in the oficial codebase parts like this :

org/apache/spark/util/HadoopFSUtils.scala

    val previousJobDescription = sc.getLocalProperty(SparkContext.SPARK_JOB_DESCRIPTION)

    val statusMap = try {
      val description = paths.size match {
        case 0 =>
          "Listing leaf files and directories 0 paths"
        case 1 =>
          s"Listing leaf files and directories for 1 path:<br/>${paths(0)}"
        case s =>
          s"Listing leaf files and directories for $s paths:<br/>${paths(0)}, ..."
      }
      sc.setJobDescription(description)

....

    finally {
      sc.setJobDescription(previousJobDescription)
    }

So my question is what is canonical way of setting jobDescription in multi-threaded application and what is a proper way of using spark sessions in this environment ?