IJ43262: SAVEBASE CAN HANG WHEN AIX_STDBUFSZ IS SET APPLIES TO AIX 7200-05

APAR status

• Closed as program error.

Error description

• with large ODM data the following might hang randomly
  export AIX_STDBUFSZ=64K
  savebase
  
  proctree will show savebase forked a child compress
  process both savebase & compress will hang :
  procstack <savebase_pid>
  6029872: savebase
  0xd01223ac  write(??, ??, ??) + 0x1cc
  0x10001c9c  IPRA.$piped_compress(??, ??) + 0x25c
  0x100029f4  IPRA.$compress_and_write(??) + 0xb4
  0x10000690  main(??, ??) + 0x150
  
  procstack <compress_pid>
  6554324: /usr/bin/compress
  0xd01223ac  write(??, ??, ??) + 0x1cc
  0xd0120bbc  _xwrite@AF19_9(??, ??, ??, ??, ??) + 0x5c
  0xd01200c8  _xflsbuf(??) + 0xc8
  0xd01207d8  __flsbuf(??, ??) + 0x98
  0x10000c10  IPRA.$output(??) + 0x330
  0x1000186c  compress() + 0x2ec
  0x10001eb8  do_stdin() + 0x2f8
  0x10002c90  main(??, ??) + 0xc30
  

Local fix

• unset AIX_STDBUFSZ before calling savebase or any thing
  that could call savebase at the end.
  

Problem summary

• Both savebase and compress can be stuck in write when the
  pipelines are full as they are in writes, none will be able
  to read to free some space in the pipelines and make the
  make the overall progress.
  proctree will show savebase forked a child compress
  process both savebase & compress will hang :
  procstack <savebase_pid>
  6029872: savebase
  0xd01223ac  write(??, ??, ??) + 0x1cc
  0x10001c9c  IPRA.$piped_compress(??, ??) + 0x25c
  0x100029f4  IPRA.$compress_and_write(??) + 0xb4
  0x10000690  main(??, ??) + 0x150
  
  procstack <compress_pid>
  6554324: /usr/bin/compress
  0xd01223ac  write(??, ??, ??) + 0x1cc
  0xd0120bbc  _xwrite@AF19_9(??, ??, ??, ??, ??) + 0x5c
  0xd01200c8  _xflsbuf(??) + 0xc8
  0xd01207d8  __flsbuf(??, ??) + 0x98
  0x10000c10  IPRA.$output(??) + 0x330
  0x1000186c  compress() + 0x2ec
  0x10001eb8  do_stdin() + 0x2f8
  0x10002c90  main(??, ??) + 0xc30
  

Problem conclusion

• Change the code to use  unblocked writes and use poll
  so that it can both read & write based on poll event
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

  IJ38959

• APAR is sysrouted TO one or more of the following:

Fix information

• Fixed component name

  AIX V7.2

• Fixed component ID

  5765CD200

Applicable component levels