Skip to content

Memory blow up when an update spec looks back more than a couple of steps into input streams relative index #27

@pt7k

Description

@pt7k
  1. Summary

    • Description: Bug Summary in 2 to 3 sentences.
    • It is possible to update an output stream using the pointwise revision mechanism with a spec that can look back at the content of input streams from before the update using t-n relative index.
    • If n is bigger than t at the time of update a special mechanism is triggered and the first n output steps after the update become unspecified and default to 0, otherwise just the initial step is 0.
    • But in both of these cases, applying an update spec with n bigger than say 3 leads to a huge memory consumption and long processing times.
  2. Environment

    • Tau Version: for now "v0.7-alpha"
    • Build Number or Date: 98f5522
    • Operating System: Ubuntu
  3. Steps to Reproduce
    Step 1: What input did you first provide?
    r o1[t]=i1[t] && u[t]=i0[t]

    Step 2: Any additional input to Tau?
    Update spec file initiating the update in step 5:

F
F
F
F
F
o1[t]=i1[t-4]

And an sbf input file:

1
1
1
0
1
1
0
0
0
0
0
0
0
0
0
0

Expected Result
Same results as can be obtained with t-2 or t-3 relative index, bar the shift in output.
Actual Result
For the run presented in the screenshots, it took some time and RAM consumption was gradually increasing reaching a few extra GB by the time it went beyond the update step.
But when I ran the same spec with just increasing the relative index to more than 4, RAM consumption's increase rate went through the roof, not just the amount but speed of increase too.
4. Additional Information

  • Error Messages or Logs:
  • I terminated the process just before running out of RAM when updating with a longer relative index
  • Screenshots:

Image

Image

Image

  1. Severity and Impact

    • Severity Level: Major
    • Impact on Work: Program Hanging/possibly crashing due to memory overconsumption
  2. Workaround (if applicable)

    • Tell us if you found a way around this problem:
    • Currently it seems that minimizing the relative length of look backs in update specs is the only way to mitigate the issue.
  3. Contact Information

    • Name: pt7k
    • Email: Telegram @ksdjfskfh
    • Additional Notes: Any other message for the Tau Team

Submission Instructions

  • Make sure to attach any input files to this bug report.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions