DRF生产环境API调用远慢于开发环境是何原因如何优化？

问题现象 #

• 部署架构：Django应用基于Docker+Nginx部署在AWS EC2实例，本地开发环境与生产环境均使用PostgreSQL数据库，两端存储数据完全一致，问题接口返回数据量仅为50KB
• 性能差异：本地环境无论是否开启DEBUG模式，该接口调用耗时稳定在600-800ms区间；生产环境同接口调用耗时达到15-20s
• 初步排查：已在所有适用场景使用select_related与prefetch_related完成数据库查询优化，进一步定位发现14-19s的耗时集中在序列化器的to_representation方法执行阶段，其余API业务代码总耗时仅1-2s

待排查问题 #

• 造成本地与生产环境巨大性能差异的核心原因是什么？
• Django在DEBUG模式与生产环境下处理API调用的逻辑存在哪些区别？
• 除数据库查询优化外，还可以通过哪些调整提升生产环境的API调用速度？
• 当前给出的序列化器与对应模型定义中，是否存在导致序列化阶段性能阻塞的问题？

关联代码 #

DataSource序列化器定义 #

class DataSourceSerializer(serializers.ModelSerializer):
    instance = InstanceSerializer()
    id = serializers.IntegerField(required=False)

    class Meta:
        model = DataSource
        exclude = ('previous', 'version', )

        extra_kwargs = {
            'password': {
                'write_only': True,
                'required': False
                },
            'db_staging': {
                'required': False
                }
        }

    def validate(self, data):
        password = data.get('password')

        if password in ['', None] and self.instance.password in ['', None]:
            raise serializers.ValidationError({"password": "This field is required"})

        return data

    # TODO: Merge in ROAD-1234
    def create(self, validated_data):
        session = self.context.get('request', None)
        if session is not None:
            session = Session.objects.get(session_key=session.session.session_key)
        instance_data = validated_data.pop('instance', None)

        # test config values
        try:
            test_connection(validated_data)
        except (exc.OperationalError, err.OperationalError):
            error = {'message': "Invalid credentials. Unable to connect to the data source."}
            raise serializers.ValidationError(error)
        except exc.TimeoutError:
            error = {'message': "Connection timed out. Unable to connect to the data source."}
            raise serializers.ValidationError(error)

        instance, _ = InstanceMetadata.objects.get_or_create(session=session, **instance_data)
        obj, _ = DataSource.objects.get_or_create(instance=instance, version=instance, **validated_data)
        return obj

DataSource模型定义 #

class DataSource(models.Model):
    id = models.AutoField(primary_key=True)
    instance = models.ForeignKey(InstanceMetadata, on_delete=models.CASCADE, related_name='instance')
    db_type = models.CharField(max_length=200)
    host = models.CharField(max_length=200)
    port = models.IntegerField()
    username = models.CharField(max_length=200)
    password = models.CharField(max_length=200)
    db_name = models.CharField(max_length=200)
    db_staging = models.CharField(max_length=200, default=None, null=True)
    version = models.ForeignKey(InstanceMetadata, on_delete=models.CASCADE, null=True)
    previous = models.ForeignKey('self', on_delete=models.SET_NULL, null=True, default=None, to_field='id')

    objects = managers.DataSourceManager()

    def natural_key(self):
        return (self.db_name, ) + self.instance.natural_key()
    natural_key.dependences = ['spinner.InstanceMetadata']

    class Meta:
        constraints = [
            models.UniqueConstraint(fields=['version', 'instance', 'db_name'],
                                    name='unique source')
        ]

    def get_unique_attr(self):
        return {
            'db_name': self.db_name,
            'instance__name': self.instance.name}

内容的提问来源于stack exchange，提问作者Judy T Raj