Hi,
[This email should be sent on September 25, 2019, but it has not been sent successfully due to Netease Master Mailbox Client.] This is indeed a problem encountered with the Dynamic Parameter Adapter(DPA) module. More work is needed to make the module better. Currently, the DPA module estimates a load of each storage group(sg) in the most pessimistic way, which is the maximum load the sg may have. In most cases, it indeed exaggerates the current load on the system. If the module tells you that the system’s load is too large, it actually means that if all sgs have maximum loads and in this case, you execute the command may crash the system. Therefore, the DPA module considers the maximum possible load rather than the real-time load of the system. which ensures that the system will not be out of memory, but to some extent sacrifices the availability of the system. It’s a trade-off between these two aspects, it’s better to develop a module that monitors the real-time writing load of the system to cooperate with the DPA module. Best Regards, ————————————————— Tianan Li School of Software, Tsinghua University 李天安 清华大学 软件学院 On 09/25/2019 14:38，Xiangdong Huang<saint...@gmail.com> wrote： Hi, The codes of iotdb-benchmark is not in our repository.. but I think it is fine to discuss the issue about it (if it can help us to improve IoTDB project). As for the issue of Dynamic Parameter Adapter, I agree with @Tian jiang. Can the author (@tianan) give more information about it? and then let's find a solution if some guys are interested in it.. Besides, when it judges the workload to be unacceptable and refuses to create time series, it does not leave enough logs or give enough feedback to the user to indicate the cause. This should be reported as a new issues on JIRA. Best, ----------------------------------- Xiangdong Huang School of Software, Tsinghua University 黄向东 清华大学 软件学院 Tian Jiang <jt2594...@163.com> 于2019年9月25日周三 上午11:49写道： Well, it turns out that the Dynamic Parameter Adapter thought the memory is insufficient according to the workload and refused to create new time series. But actually when I turned it off, I found that was a misjudge since it is okay using the default parameters. I think the definition of "workload too heavy" in the Dynamic Parameter Adapter is too strict, which dramatically underestimates the ability of the system. Besides, when it judges the workload to be unacceptable and refuses to create time series, it does not leave enough logs or give enough feedback to the user to indicate the cause. | | Tian Jiang | | jt2594...@163.com | 签名由网易邮箱大师定制 On 9/25/2019 11:31，徐毅<xuyith...@126.com> wrote： Hi, You can contact liurui directly. I guess there are some error configurations in your file. Thanks XuYi On 09/25/2019 11:04，Tian Jiang<jt2594...@163.com> wrote： Hi, When I used IoTDB-benchmark to perform insertions to IoTDB (both are the master branch), I got some really weird phenomenons. 1. This is my settings of the benchmark and I expected to have 14 threads to perform the insertion, while actually I only found 2 threads working. 2. I got a lot of failures when creating new time series, and the error code did not explain it well. Since I started the experiment with a clean IoTDB, I do not think this was caused by creating existing timeseries and I did not find related logs on the server side. 3. This may be related to issue2. In the first attempts of insertions, I got some errors indicating that the time series are not created successfully. These issues have blocked my progress of testing TTL, I hope someone could fix it ASAP. | | Tian Jiang | | jt2594...@163.com | 签名由网易邮箱大师定制
