[BUG-202968] [Alex Ivy] Pipelining still blows up frequently resulting in corrupted textures & mesh.

[sl-service-account]

Steps to Reproduce

• Teleport about different regions on Second Life Release 5.1.0.511732.

• Observe how textures & mesh render.

  Observed Behaviour

• Textures & mesh still render corrupted fairly frequently on the Alex Ivy release 5.1.0.511732

• This is basically the same problem as BUG-41295 with pipelining blowing up, apart from the resulting crash is now fixed.

• Clearing viewer cache does not fix the problem, infact it's more likely to reproduce when fetching new textures.

• See linked Demo Video 1 showing corrupted rainbow textures & corrupted mesh that's rendering as vertex vomit.

• Attached Whirly_1.log. Note the usual problems showing pipelining is getting out of sync.

  
  2018-01-22T09:56:00Z WARNING: #Texture LLTextureFetchWorker::callbackHttpGet: data_size = 937 > requested: 600
  2018-01-22T09:56:00Z WARNING: #Texture LLTextureFetchWorker::doWork: Partial HTTP response produces break in image data for texture 26e39de6-7a8f-f425-27fc-c88c4522b817.  Aborting load.
  2018-01-22T09:56:01Z WARNING: #LLMediaDataClient LLMediaDataClient::serviceQueue: Could not send request request: num=18 type=GET ID=79825324-c305-be11-254f-2eff5a0a5b94 face=-1 #retries=1 (empty cap url), will retry.
  2018-01-22T09:56:01Z INFO: LLKDUMessage::put_text: KDU Error: Kakadu Core Error:
  
  2018-01-22T09:56:01Z INFO: LLKDUMessage::put_text: KDU Error: Missing or out-of-sequence tile-parts for tile number 
  2018-01-22T09:56:01Z INFO: LLKDUMessage::put_text: KDU Error: 0
  2018-01-22T09:56:01Z INFO: LLKDUMessage::put_text: KDU Error:  in code-stream!
  2018-01-22T09:56:01Z WARNING: #Texture LLTextureFetchWorker::callbackDecoded: DECODE FAILED: 34f0c099-de3a-4e71-231e-4ba1e8f846b1 Discard: 0
  2018-01-22T09:56:01Z WARNING: #Texture LLTextureFetchWorker::callbackHttpGet: data_size = 2713 > requested: 600
  2018-01-22T09:56:01Z WARNING: #Texture LLTextureFetchWorker::doWork: Partial HTTP response produces break in image data for texture 492829ed-e04e-7b9e-5199-509914a52110.  Aborting load.
  2018-01-22T09:56:01Z WARNING: #Texture LLTextureFetchWorker::callbackHttpGet: data_size = 1140 > requested: 600
  2018-01-22T09:56:01Z WARNING: #Texture LLTextureFetchWorker::doWork: Partial HTTP response produces break in image data for texture 74d306a7-4117-413f-6cbe-5b78723cfb75.  Aborting load.
  2018-01-22T09:56:01Z WARNING: LLViewerFetchedTexture::updateFetch: 74d306a7-4117-413f-6cbe-5b78723cfb75 Fetch failure, setting as missing, decode_priority 2300019.500000 mRawDiscardLevel 32767 current_discard -1 stats 00ce0000
  2018-01-22T09:56:01Z WARNING: LLViewerFetchedTexture::setIsMissingAsset: 74d306a7-4117-413f-6cbe-5b78723cfb75: Marking image as missing
  2018-01-22T09:56:01Z WARNING: #Texture LLTextureFetchWorker::doWork: Partial HTTP response produces break in image data for texture 727b0e5f-eab3-594f-d5cc-69aa84e2c56b.  Aborting load.
  2018-01-22T09:56:02Z WARNING: #LLMediaDataClient LLMediaDataClient::serviceQueue: Could not send request request: num=18 type=GET ID=79825324-c305-be11-254f-2eff5a0a5b94 face=-1 #retries=2 (empty cap url), will retry.
  2018-01-22T09:56:02Z WARNING: #Texture LLTextureFetchWorker::callbackHttpGet: data_size = 1449 > requested: 600
  2018-01-22T09:56:02Z WARNING: #Texture LLTextureFetchWorker::doWork: Partial HTTP response produces break in image data for texture 848648c5-6939-7abf-465e-a2ecfe030ca9.  Aborting load.
  2018-01-22T09:56:02Z WARNING: #Texture LLTextureFetchWorker::callbackHttpGet: data_size = 2666 > requested: 1449
  2018-01-22T09:56:02Z WARNING: #Texture LLTextureFetchWorker::doWork: Partial HTTP response produces break in image data for texture 9b1fa4db-f91e-09b6-a94c-1f4414c4a15c.  Aborting load.

  Expected Behaviour

  Pipelining to work ok like it does on pre-Alex Ivy viewers.

  Workaround

  Disable HttpPipelining.

Attachments

• Whirly_1.log 2018-01-22 05:12:36

Links

Related

• [ ] https://github.com/secondlife/jira-archive/issues/12846
• [ ] https://github.com/secondlife/jira-archive/issues/3191

  Duplicates

• [ ] https://github.com/secondlife/jira-archive/issues/3636
• [ ] https://github.com/secondlife/jira-archive/issues/3547

Original Jira Fields

| Field | Value | | ------------- | ------------- | | Issue | BUG-202968 | | Summary | [Alex Ivy] Pipelining still blows up frequently resulting in corrupted textures & mesh. | | Type | Bug | | Priority | Unset | | Status | Accepted | | Resolution | Accepted | | Reporter | Whirly Fizzle (whirly.fizzle) | | Created at | 2018-01-22T10:03:23Z | | Updated at | 2018-06-07T14:18:09Z | ``` { 'Business Unit': ['Platform'], 'Date of First Response': '2018-01-22T04:47:40.272-0600', 'ReOpened Count': 0.0, 'Severity': 'Unset', 'System': 'SL Viewer', 'Target Viewer Version': 'viewer-development', 'What just happened?': 'Filling in...', 'What were you doing when it happened?': '....', 'What were you expecting to happen instead?': '....', } ```

[sl-service-account]

Ansariel Hiller commented at 2018-01-22T10:47:40Z

I can observe this as well, especially on the first login after completely wiping the cache inventory (texture, object & inventory cache). There is a high chance the pipelining issue is going to occur then. I can provide a log if desired, but the actual log output is the same as Whirly's already shows: Some response seems to get lost and as a result and the responses coming in are processed wrong. This is because the viewer waits for a particular response, but due to a response getting lost, it is going to process the response for the next request. To make it more obvious:

Viewer requests:

Req-ID       Size
======     ======
     1        555
     2      12345
     3       4567
     4         10
     5       9999

Viewer processing response coming in (assuming response for request 2 got lost):

Res No.    Expected Req-ID    Expected Size    Actual Req-ID    Actual Size
=======    ===============    =============    =============    ===========
      1                  1              555                1            555
      2                  2            12345                3           4567
      3                  3             4567                4             10
      4                  4               10                5           9999

This raises the question why HTTP pipelining is still pursued anyway. Google Chrome and Mozilla Firefox for instance even removed HTTP pipelining support because it's rather fragile and error-prone, not to mention it can cause head-of-line blocking and resulting in a performance degradation.