java - 如何解决scala代码中超出GC过载限制的问题

Question

我只是从 solr 节点读取记录。我的代码只读取给定日期范围内的记录。我检查过，它适用于 50K 记录，但我尝试了 100k，然后发现超出了 GC 过载限制。

我的代码在scala中是这样的:

def querySolr(core: String, selectQuery: String, server: SolrClient,
            pageNum: Int, pageStart: Int, pageSize: Int): (Long, SolrDocumentList) = {
    val query = new SolrQuery(core)
    query.setQuery(selectQuery)
    query.setStart(pageStart)
    query.setRows(pageSize)
    val response: QueryResponse = server.query(query)
    val results: SolrDocumentList = response.getResults
    val total = results.getNumFound
    (total, results)
  }

  def pageCalc(page: Int, pageSize: Int, totalItems: Long): (Int, Long, Long) = {
     val from = ((page - 1) * pageSize) + 1
     val to = totalItems min (from + pageSize - 1)
     val totalPages = (totalItems / pageSize) + (if (totalItems % pageSize > 0) 1 else 0)
     (from, to, totalPages)
  }

 def getRecordsFromSolr(core: String, solrhost: String, userName: String, password: String,
                     query: String): List[SolrDocument] = {

     val startTime = System.nanoTime()
     val url = "https://" + solrhost + ":8983/solr/" + core
     val solrPort = 8983

     val builder: SSLContextBuilder = new SSLContextBuilder()
     builder.loadTrustMaterial(null, new TrustSelfSignedStrategy())
     val sslsf: SSLConnectionSocketFactory = new SSLConnectionSocketFactory(
      builder.build(),  SSLConnectionSocketFactory.ALLOW_ALL_HOSTNAME_VERIFIER
)

   val credsProvider: CredentialsProvider = new BasicCredentialsProvider()
credsProvider.setCredentials(
    new AuthScope(solrhost, solrPort),
   new UsernamePasswordCredentials(userName, password))

   val httpclient: CloseableHttpClient =HttpClients.custom().setSSLSocketFactory(sslsf).setDefaultCredentialsProvider(credsProvider).build()

  val server: SolrClient = new HttpSolrClient(url, httpclient)

   logger.info("solr connection completed")

   val pageSize = 1000
   var pageNum = 1
   var nextPage: (Int, Long, Long) = (0, 1000, 0)
   var offset: Long = 0

   var totalResult = querySolr(core, query, server, pageNum, 0, pageSize)
   var total = totalResult._1
   var results: List[SolrDocument] = totalResult._2.toList
  while (total > offset) {
    offset += pageSize
    pageNum += 1
    nextPage = pageCalc(pageNum, pageSize, total)
    totalResult = querySolr(core, query, server, pageNum, nextPage._1, pageSize)
    total = totalResult._1
    results = (results ++ totalResult._2.toList)
 }
 }

java.lang.OutOfMemoryError:超出 GC 开销限制

如何避免内存泄漏。我尝试过每个核心 8GB。并且表包含数百万条记录。

我发现 60K 记录出现以下错误:

Exception in thread "main" org.apache.spark.SparkException: Job aborted due to stage failure: Serialized task 0:0 was 18311053 bytes, which exceeds max allowed: spark.akka.frameSize (10485760 bytes) - reserved (204800 bytes). Consider increasing spark.akka.frameSize or using broadcast variables for large values.

Answer 1

读取太大的 solr 响应时，通常会出现 OutOfMemoryError 错误。

因此解决方案是最小化 solr 响应:

1. 限制行大小
2. 限制返回的字段列表(参数fl)。特别是包含大型索引文档(例如 pdf)的字段可能会增长到很大。

如果这没有帮助，我建议分析您的 solr 响应。尝试找出实际的 solr 查询并在浏览器中执行它。