Beta 260/262 hangs

[vbn]

For the last few days, HFS hangs under high load. When I wake up its hung with Tray Icon showing 450 connections and climbing and not reducing at all, like HFS just stops responding to them but keeps receiving.

I have high usage during US days (server stays contant at around 80-90 Mbps) and thats when this has started happening. I have set connection timeout to 10 secs in an effort to kill dormant ones but didnt help. Even set limits of connections to 300 but no use.

Any ideas on how to properly debug this situation ?

[rejetto]

450 connections at the same time?
As a test, try to disable "enable recursive listing" (in menu > virtual file system).
This is one of the heavier things in the server, and is used for "search" and "archive" (that will stop working).
See if the problem is bound to this.

[vbn]

Ok...will monitor couple of days now ...also Persistent Connections somehow had become unchecked...could that be the root cause ?

[rejetto]

it affects performance, but "how" it depends on the kind of traffic

[vbn]

OK...I think HFS is not able to handle high loads... I have set the tray icon to show Number of connections and when it dies, the tray icon count keeps rising and rising...see image

I have tried setting limits etc but its not working. When its stuck and you try to open HFS it opens the Window frame and nothing inside. Right click menu opens a blank menu without options. It stops responding totally. Only way is to kill task and restart it.

File types are zip/rar with size range 80MB to 200MB and I get high load !

[TSG]

I have had this too, in latest beta, testing locally, HFS just locks up and wont respond to any requests. I actually have to close HFS and restart it.

Usually happens after I try to upload something big, or download something big with multiple connections (stream a large video for example). DivX Web Player for example will use one connection, but constantly connect and disconnect. Instead of a steady file download, if I persist with that, HFS will hang.

[vbn]

I narrowed it down to the following scenario :

There is a newest development on Android side for HTC HD2 over at xda-developers and when a Chef releases new ROM, people go crazy worldwide. This sudden rush of connections (if HFS runs without Limits) , crashes it.

I had to start a seperate instance of HFS just for serving these Android folders and havent had crash in last day or so.

How can we debug this better application wise ? I am ready to fully cooperate with rejetto and provide RDP access to the server if he wants to take a look.

[rejetto]

1. can you tell how did the "enable recursive listing" affect the problem?
2. what can you say about the load the "android dedicated server" is getting? some number?
3. try removing the "archivable" flag for the home/root folder, and see if this affects in any way.

[vbn]

1. can you tell how did the "enable recursive listing" affect the problem?

I have disabled recursive listing on both instances but had not helped

2. what can you say about the load the "android dedicated server" is getting? some number?

The average number of connections is around 25 constantly. Peak time it goes around 90+.
Average hourly data transfer on that IP (as per reports from my DC) ranges 7GB+ with filesizes in range of 80-200 MB.

3. try removing the "archivable" flag for the home/root folder, and see if this affects in any way.

Archivable was already disabled when this started happening

-------------------------

Could it be because of the app log and graph running and causing crash on high number of connections. Also just saw a *dump-bin file in HFS directory last modified on 15th Aug....its about 10 MB...I can send you the link for that if it helps

[rejetto]

that is a dump of the network traffic, you probably enabled it by chance. Trash it.
High load servers should consider logging only to file, by disabling "log to screen".
anyway you are correct, HFS was not designed for high loads, it focuses on usability.
another try is to set a limit on contemporary downloads. Setting 50 or 100 could be a way to also improve quality of service.