| Summary: | [scale] - loading XHR objects taking too long on landing page on WAN, with many objects (525 hosts, 11071 VMs) | ||
|---|---|---|---|
| Product: | [oVirt] ovirt-engine | Reporter: | Eldad Marciano <emarcian> |
| Component: | Frontend.WebAdmin | Assignee: | Greg Sheremeta <gshereme> |
| Status: | CLOSED CURRENTRELEASE | QA Contact: | eberman |
| Severity: | medium | Docs Contact: | |
| Priority: | unspecified | ||
| Version: | 3.6.2 | CC: | bugs, dagur, eberman, vszocs, ykaul |
| Target Milestone: | ovirt-4.2.0 | Keywords: | Performance |
| Target Release: | 4.2.0 | Flags: | rule-engine:
ovirt-4.2+
rule-engine: planning_ack+ rule-engine: devel_ack+ eberman: testing_ack+ |
| Hardware: | x86_64 | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | Bug Fix | |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2018-01-12 12:56:39 UTC | Type: | Bug |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | UX | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
|
Description
Eldad Marciano
2016-01-25 11:17:25 UTC
Eldad, - Does this happen all the time? I assume only on high scale environment, right? - What are the numbers of objects that you have in the system? - What browser is used? did you try with different browsers? - Is the engine and/or the client maxed out, CPU or memory-wise? - Is the engine and the UI on the same LAN (at least on the same site) ? (In reply to Yaniv Kaul from comment #1) > Eldad, > - Does this happen all the time? I assume only on high scale environment, > right? yes, specially on high scale. > - What are the numbers of objects that you have in the system? DC 2, Clusters 3, SD 13, hosts 525, vms 11071. > - What browser is used? did you try with different browsers? firefox and chrome - firefox working much better, but still slow as i mention above. > - Is the engine and/or the client maxed out, CPU or memory-wise? engine is pretty idle, but the client running the browser shows high performance. > - Is the engine and the UI on the same LAN (at least on the same site) ? No - WAN, we need compare it to LAN. Hi Eldad, > see the following link > http://pastebin.test.redhat.com/343246 Those are GWT RPC request payloads. Each request specifies the remote interface (GenericApiGWTService), specific method of that interface (runMultipleQueries etc) along with any method parameters. As you wrote, GWT RPC requests are issued via XMLHttpRequest (XHR). Obviously, HTTP over WAN has dramatic impact on response times. When GWT RPC response arrives on client, it is evaluated which can take up some more time (bigger response payloads take more time to process). So the response times you've recorded are essentially a combination of high network latency + big response payload (lots of VMs etc). Unfortunately, we can't do much about GWT RPC response processing since that code is part of GWT SDK itself. We could attempt to "hack" it but I would strongly object to that simply due to the risk involved. What we can do is to reduce/optimize client's requests in environments like WAN + scale combo. For example, have WebAdmin fetch and display a limited amount of data, which should make it more responsive. > more over seems like some XHR (GWT) queries repeat them self. In WebAdmin, each main tab grid automatically refreshes its data. This is by design as we currently don't have any mechanism to push data from server to client. > it takes something like ~10min to completely load the page. There are lots of factors involved behind this number: * specific browser (rendering engine + JS engine performance) * network latency (LAN vs WAN etc) to fetch remote resources - including any HTML/JS/CSS/etc fetched beyond initial load - including any GWT RPC requests * amount of initial data to display (GWT RPC response processing) Some things we can't influence, but we can make optimizations: 1, reduce amount of data displayed in high latency / scale env. - UI grids showing less data rows (to be made configurable) 2, reduce number of necessary HTTP roundtrips - don't refresh UI grids too often (configurable already) - inline GWT selector script inside initial page What are your thoughts? @Eldad -- please retest and verify / fail, or close currentrelease. Please see comment #2 - it was reported on WAN, not LAN. Did you test it with some latency / bandwidth constraints (I personally believe latency is the main factor here, not the bandwidth!)? Removing Need Info as this bug is already closed. |