ma.log filling up with NoSQL Batch Writer Fatal Error

[cmusselm]

@CraigWeb Thanks for the reply. I don't think that's the case. We only start the app via a Windows service which runs as Local System account.

The fact that each time this happens, we found a scripting data source that hits the /data-sources/status endpoint with the error seemed odd, and made us wonder if that endpoint could be contributing to the error.

Is there any way to pull that data differently? It doesn't seem to be set up like a normal data point. If we could reference it's history (More than the previous 10) we may be able to bypass the scripting DS that we currently use to get this info.

Thanks,
Chad

[terrypacker]

@cmusselm a few thoughts for you.

• You might be able to get this information via setting up the Internal Data Source to have a point that tracks what you are looking for. However I don't easily see how the status endpoint being hit will affect the shard files...
• Did you triple check your antivirus software, it might be locking that file?

[cmusselm]

@terrypacker I'll check again with our infrastructure team to ensure that all antivirus software has the DB directories excluded, and that no other software could be hitting them.

For the internal data points, I have posted a couple questions about them in 2020, but based on the answers, they don't work the way we would need for our client.
https://forum.mango-os.com/topic/4925/poll-success-percentage-over-what-timeframe

https://forum.mango-os.com/topic/4948/question-on-internal-metric-previous-sequential-successful-polls

I wouldn't think that this endpoint is causing the issue, but it seems odd that every time this happens, a Data Source calling that endpoint is generating the error. Do you know where the Data Source Runtime Status data is kept? Is it somewhere in the NoSQL file structure, and is there a way that we could tell Mango to keep more than the previous 10 data point values? If we could get it to store the previous 2 - 4 weeks, then we could probably use the status endpoint and not need a new data source to store its data for longer periods.

Thanks,
Chad

[terrypacker]

@cmusselm

Do you know where the Data Source Runtime Status data is kept?

It's only in memory (cached on the running data source).

and is there a way that we could tell Mango to keep more than the previous 10 data point values?

This is hard coded to keep the last 10 poll statistics only.

Just to be sure I read through the status endpoint code and I can't see how it would affect a shard. So either this is an impressive coincidence or we are missing something.

[cmusselm]

For tracking purposes, I wanted to let you know that we got another corrupt file a night or two ago. It again was from one of our Runtime scripts. I went through every one and clicked the validate button, and sure enough one complained about the file.

I've run out of ideas on what could be locking the file outside of Mango. I'll ask if our infrastructure team can do another sweep of the system to see if anything sticks out.

Thanks,
Chad

[cmusselm]

We've now had 3 more files get corrupted, all are trying to be written to by our scripting datasources that hit the /data-source/status endpoint.

Is there a way to get this looked into more closely? Should I open a support ticket?

100% of these errors have been with datasources we use to hit the status endpoint.

ERROR 2022-08-30T17:12:49,279 [high-pool-2-thread-159790 --> Polling Data Source: datasource_name_here-runtime_status] - Uncaught Task Exception 
com.serotonin.ShouldNeverHappenException: java.lang.RuntimeException: java.io.FileNotFoundException: C:\Databases\mangoTSDB\3\41918\773.data.rev (Access is denied)

ERROR 2022-08-30T17:12:52,544 [high-pool-2-thread-159645 --> Polling Data Source: datasource_name_here-runtime_status] - Uncaught Task Exception 
com.serotonin.ShouldNeverHappenException: java.lang.RuntimeException: java.io.FileNotFoundException: C:\Databases\mangoTSDB\54\33945\773.data.rev (Access is denied)

ERROR 2022-08-30T17:12:56,294 [high-pool-2-thread-159759 --> Polling Data Source: datasource_name_here-runtime_status] - Uncaught Task Exception 
com.serotonin.ShouldNeverHappenException: java.lang.RuntimeException: java.io.FileNotFoundException: C:\Databases\mangoTSDB\78\23157\773.data.rev (Access is denied)

Thanks,
Chad

[MattFox]

@cmusselm Push for a support ticket.
It's the only way to get ahold of them now.

[cmusselm]

@MattFox You're probably right. I'll do that here soon. As of now, I'm trying to re-create the issue in our test environment. We have 150+ scripts that run every 6 seconds to get the runtime info for various data sources in prod.

Hopefully doing this in test will yield similar issues and can help with additional troubleshooting.

[cmusselm]

One thing we're doing that could be contributing to this is using .set inside a for loop when we have pulled multiple runtime records.

        for (var i = noRec; i > 0; i--) {
            var pollDuration = content.latestPolls[successLen-i].duration/1000;
            var pollTime = content.latestPolls[successLen-i].startTime;
            varPollDuration.set(pollDuration, epoch(pollTime));
        } //End for

We're going to replace this with an API call the /point-values to see if pushing the array of values and time to the data point will help resolve this.

I'll post an update when done and tested out.

[MattFox]

@cmusselm talking from experience, if you start hammering the api, you can get dashboard connection issues.
Using the java calls is far better as it is more direct.
Perhaps try inserting a RuntimeManager.sleep() call instead?
Space out each write with ten milliseconds to give your system enough time between writes.
That and perhaps increase the cache size of those points to ten if you have not done so already.

Fox

[cmusselm]

@MattFox Thanks for the insight. I'll definitely explore that and let you know what happens.

Chad

[cmusselm]

Quick update on my progress.

It took a little longer that desired, but we were finally able to get the corrupt files cleared out, and all of our Runtime DS Scripts updated with the sleep that @MattFox suggested, as well as caching 10 DPs for those points.

Hopefully all is good, and we don't see corrupt files any longer. I'll give it a month or so and post an update to hopefully close this out once and for all.

Thanks,
Chad

[MattFox]

@cmusselm Good Luck!

Fox

[cmusselm]

It's been well over a month and we haven't seen the issue come back. I think it's safe to say the problem is resolved.

Thanks @MattFox for the suggestions and being active in the forum! Your posts have helped me more than you know.

[MattFox]

@cmusselm Thanks for letting me know!