Bad request

[msageryd]

You're right. I see Setting http-server strategy for rendering in the log when jsreport starts.

My config looks like this:

{
  "httpPort": 5488,
  "allowLocalFilesAccess": true,
  "store": {
    "provider": "fs"
  },
  "blobStorage": {
    "provider": "fs"
  },
  "templatingEngines": {
    "numberOfWorkers": 2,
    "timeout": 300000,
    "strategy": "in-process"
  },
  "chrome": {
    "timeout": 300000
  },
  "extensions": {
    "scripts": {"timeout": 300000},
    "chrome-pdf": {
      "strategy": "chrome-pool",
      "numberOfWorkers": 3
    },
    "express": {
      "inputRequestLimit": "50mb"
    },
    "child-templates": {
      "parallelLimit": 5
    }
  }
}

Can you spot any error?
I tried to deliberately corrupt the config file to ensure that it's actually loaded. When I do this, the server complains about a bad config file and does not start.

Can these settings be set anywhere else and override the config file?

[msageryd]

I read the startup log a bit more thorough.
What does "/app/data/data: File exists" mean? I don't think I have copied anything to that folder.

Here is my Dockerfile

FROM jsreport/jsreport:2.6.0

COPY license-key.txt /app
COPY ./mounted/jsreport.config.json /app
COPY ./mounted/data /app/data

RUN cd /app && npm install \
  moment \
  moment-timezone \
  intl \
  dd-trace \
  handlebars-helpers \
  node-fetch \
  jsreport-assets \
  jsreport-static-pdf

[msageryd]

I suppose that ENV-params have higher priority than the config-file?
In that case I think it's impossible to change strategy with your official docker image. I found this in your Dockerfile:

ENV templatingEngines:strategy http-server

[jan_blaha]

Hm. Ok. I forgot about that. I am very sorry. This is badly set a default value.

You need to add this to your dockerfile

ENV templatingEngines:strategy in-process

We will think through how to fix this.

[jan_blaha]

I link here github issue where this will be solved
https://github.com/jsreport/jsreport/issues/643

I apologize again for your troubles.

[msageryd]

Ok, I'll set it in my Dockerfile.

I'm getting odd results when I build my production image. In order to be able to run jsreport on Fargate I need to copy all templates into the image at build time. But as soon as I do this the container no longer maps my local volume. I'll try to explain:

Scenario 1:

1. Build image without copying templates to /app/data
2. Start local docker image with docker-compose and map local volume to /jsreport
3. All changes I do in Studio is now persisted in my local folder which I can version control

Scenario 2:

1. Build an image and copy templates (COPY ./mounted/data /app/data in Dockerfile)
2. Push to production at AWS
3. jsreport server at AWS now has my templates embedded
4. Start local container via docker-compose (exactly as in scenario 1)
5. Changes i Studio is not persisted in my local folder

How are we supposed to use this concept? I can see in the official docker image that these two scenarios are taken care of, i.e. you check is /jsreport is mounted and in that case link /app/data to this local folder. But I'm not sure that it works as intended.

In particular, I'm wondering why you have VOLUME ["/jsreport"] in your Dockerfile. Wouldn't this direct the ln commands to the internal /app/data instead of the locally mounted folder? I'm not very knowledgable in Docker, so this "hunch" might be completely wrong.

[msageryd]

After switching to "in-process" my heavy report works without error. I think it's more to the problem than just bad error reporting. The "Bad request" came after about 4 seconds. Efter switching to in-process the report is rendered in just over a minute.

I do wonder if in-process rendering makes the server more sensitive to overload? After switching to in-process and deliberately overloaded jsreport (10 heavy reports in parallell) the server crashed. I haven't made a side by side comparison, byt my feeling is that before switching to in-process, jsreport could manage more reports in parallell. They took a long time to render, but it didn't crash.

I have now adjusted my queue-handling so I have a separate queue for heavy reports. This queue only puts through one report concurrently. Is this how I have to solve this? I thought that jsreport could handle parallel requests.

[jan_blaha]

in-process runs js evaluation inside the main process. This means that running many of the requests with heavy memory consumption in parallel can cause OOM.
You should use a properly configured web server in front jsreport, or increase node process memory limits in such case.

[msageryd]

Thanks, I'll increase max-old-memory. What do you mean by having a web server in front of jsreport? My jsreport is not available from the outside. All requests comes from another server which serves/distributes my queue jobs. One queue is for report rendering. Why would I need a web server in front of jsreport?

[jan_blaha]

No, in this case you are able to handle the traffic on your client-side where you distribute queue jobs.