error: commlib error: got read error (closing "rdXXXX/shepherd_ijs/1")

[eddiewang927]

🐛 Problem Description

When running an EDA tool that produces large and high‑frequency stdout output under an OCS interactive job environment (e.g., qrsh), the session consistently terminates with the following error:

error: commlib error: got read error (closing "rdXXXX/shepherd_ijs/1")

To verify whether this is caused by stdout pressure, I created a simplified script that simulates the EDA tool’s high‑volume output behavior. The same error occurs reliably.

---

🔁 Reproduction Steps

Run command

qrsh -now n -verbose -cwd test.sh

test.sh content (simulating heavy EDA stdout)

#!/bin/bash

stdbuf -o0 -e0 yes "$(cat <<'EOF'

VDD = 0.4998 V ; VPP = 0.4998 V ; C1 = 0 F ; R = 0 Ohm ; C2 = 2.882e-13 F ; Slew1 = 9.0375e-10 S ; Slew2 = 9.0375e-10 S ;

      state = output_rise ; vector = A1&!A2&!A3&!A4 ; active_input = A1 ; active_output = Z ;

      pin               peak             area            width

      VDD               8.80e-05  A      1.40e-13 C      5.55e-09 S

      VPP               -5.79e-07 A      1.36e-16 C      3.80e-09 S

      VSS               8.80e-05  A      1.41e-13 C      5.55e-09 S

      VBB               4.38e-07  A      9.02e-17 C      4.08e-09 S

      state = output_fall ; vector = A1&!A2&!A3&!A4 ; active_input = A1 ; active_output = Z ;

      pin               peak             area            width

      VDD               4.54e-06  A      2.30e-15 C      4.37e-09 S

      VPP               4.57e-07 A      1.40e-16 C      3.93e-09 S

      VSS               5.20e-06  A      1.82e-15 C      4.37e-09 S

      VBB               -3.75e-07 A      9.05e-17 C      4.19e-09 S

EOF

)" | head -n 1000000

Actual Result

Within a few seconds, the interactive job fails with:

error: commlib error: got read error (closing "rdXXXX/shepherd_ijs/1")

This matches exactly the behavior observed when running EDA tools.

---

❓ Question

Is this behavior a known limitation of OCS’s IJS (Interactive Job Stream) when handling high‑volume stdout, or should it be considered a bug?

Thank you.

[jgabler-hpc]

Hi Elmo,

thank you for reporting this issue.

We consider it as a bug - qrsh needs to be able to process any amount of data through stdin/stdout/stderr.

I can reproduce it and will work on it for a next patch.

Best regards,

Joachim