You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@linkis.apache.org by rita <ri...@163.com> on 2022/08/09 03:40:42 UTC
[DISCUSS]Gateway module is faulty
Dear:
The chat records of WeChat group“Apache linkis Community Development Group”are as follows微信群"Apache Linkis 社区开发群"的聊天记录如下
————— 2022-8-6 —————
Xi 10:50
In our case, the Gateway module occasionally fails directly. Have you ever encountered this @peacewong@WDS 和平兄,我们这里gateway模块偶尔会直接挂掉,你们有遇到过吗
Mr flash闪电先生 10:56
I do find that Eureka dies a lot, solo我倒是发现eureka经常死,单机的话
Mr flash闪电先生 10:57
This could be improved by deploying two gateways如果部署二个gateway 可能会有所改善
peacewong@WDS 11:11
Did you see any reason to die?有看到什么原因挂掉吗?是不是内存设置太小。我们这边没有挂掉过
Xi 11:14
Yeah, it's OOM恩,看了下是oom了
peacewong@WDS 11:15
How much do I set it to设置的多少呢
Xi 11:15
How much do I set it to设置的是2G
peacewong@WDS 11:15
Eureka wouldn't have been normal eitherEureka正常应该也不会的
Xi 11:15
I went up to 6 grams我加到6G了
peacewong@WDS 11:16
I went up to 6 gramsdump Our 2G never failed看看,哪个比较多?我们这边2g没挂过
Xi 11:16
I am looking at the background log here for other services accessing Eureka frequently times out我这里看后台日志其他服务访问eureka经常超时
peacewong@WDS 11:16
Not normally. Eureka doesn't have a lot of requests正常不会,Eureka的请求数不多
Xi 11:18
WDS. Linkis.gateway. url= Is this also comma separated ?gateway 部署多个的时候 wds.linkis.gateway.url= 这个也是逗号分隔吗
Xi 11:24
WDS. Linkis.gateway. url= Is this also comma separated没共享这个问题有解决办法吗
Xi 11:25
Is it still possible to only point to one of the gateways when logging in with scripts用scripts登陆的时候是不是还是只能指到其中一个gateway上
Xi 11:26
Can I only point to one of the gateways when logging in the Linkis background登陆linkis后台也是只能指向其中一台gateway吗
Xi 11:26
@夏晨 @casion
Heisenbao 海森堡 11:26
If you want to share, you probably need to add a public cache想要共享的话 估计需要加公共缓存啦
Heisenbao 海森堡 11:28
The public cache is used to store the login status of the user. The login status is saved by each gateway and is not shared to other gateways.公共缓存来保存用户登录状态,现在的登录状态是由每一个gateway 保存的,没有共享到别的gateway,
peacewong@WDS 11:28
This way you don't need a public cache这种方式就不需要公共缓存
Heisenbao 海森堡 11:29
As I recall, Gateway records the login status of the user我记得gateway 会记录用户的登录状态
Xi 11:30
I try 我试下
casion 11:32
Well, using Nginx's IP hash as the load ensures that requests are always sent to a fixed gateway without changing the IP address of the same user.嗯嗯,用nginx的ip hash来做负载,能保证同一个用户ip不变前提下,始终请求到固定某台gateway上。
casion 11:34
This is 6G, but there is something wrong with the OOM. Please check it. The Gateway processing logic is relatively light, so it should not need too much memory这个6G了,还oom应该是哪里有点问题,可以排查下,gateway处理逻辑体量比较轻,不应该需要太多内存
Xi 11:38
This is when the kill task, report an error, not clear if there is a relationship, feel like a code logic problem这个是kill任务的时候,报个错,不清楚是不是有关系,感觉像是代码逻辑问题
Xi 11:39
Xi 11:39
/api/rest_j/v1/entrance/exec_id018017linkis-cg-entrance172.21.32.17:9104IDE_eps_dw_spark_2159/kill
Xi 11:39
There is a consistent error in the kill task. It looks like there is a problem with execidkill 任务的时候这里一致报错,看着execid好像在这里有问题
Xi 11:40
This is 2G error, I added to 6G restart, again observe这个是2G报错的,我加到6G重启了,再观察下
peacewong@WDS 11:43
This is not OOM, looks like the parameter parsing error?这个不是oom,看起来是传的参数解析错误了?
Xi 12:12
Xi 12:12
OOM in this OOM的在这
Xi 12:13
OOM in this刚那个是大量warn,看着像是kill任务的时候报的
peacewong@WDS 12:18
Take a look at GC logs and dump看看gc日志和dump看看