hold capture fails on unique index constraint

[Chris Sharp]

Evergreen 2.8.2 introduced a unique index on action.hold_request for current_copy (see bug 902255). We are seeing occasional errors in the logs where the application layer is trying to insert a duplicate row. Here is the client-side error message:

Tue Sep 27 2016 11:52:16 GMT-0400 (Eastern Daylight Time)

Check In Failed (in circ.util.checkin) (3):

{
 "ilsevent":"2001", "textcode":"DATABASE_UPDATE_FAILED",
 "debug":"Exception: OpenSRF::DomainObject::oilsMethodException
 2016-09-27T11:51:49 OpenILS::Utils::CStoreEditor
 /usr/local/share/perl/5.18.2/OpenILS/Utils/CStoreEditor.pm:847 <500> Error in updating a row -- please see the error log for more details\n",
 "desc":"The attempt to write to the DB failed",
 "stacktrace":"/usr/local/share/perl/5.18.2/OpenILS/Utils/CStoreEditor.pm:847 (eval 2729):1
 /usr/local/share/perl/5.18.2/OpenILS/Application/Circ/Circulate.pm:2895",
  "servertime":"Tue Sep 27 11:51:49 2016",
  "payload":
  {
   "__c":"ahr",
   "__p":
   [
    null,
    null,
    "now",
    14709465,
    "t",
    null,
    null,
    null,
    null,
    "C",
    null,
    15730851,
    "<phone number redacted>",
    null,
    null,
    170,
    "2016-09-26T00:00:00-0400",
    170,
    "2016-07-11T13:55:53-0400",
    369968,
    0,
    170,
    14709465,
    2121201,
    null,
    null,
    null,
    null,
    null,
    null,
    "f",
    null,
    null,
    null,
    null,
    null,
    "t",
    null,
    null,
    null,
    "f"
   ]
  }
  ,
  "pid":27220,
  "message":"DATABASE_UPDATE_FAILED",
  "route_to":" "
}

Attaching scrubbed-of-sensitive-data server-side logs below.

[Bill Erickson (berick)]

I see this often as well. Have not yet tried to find the source of the problem. FWIW, I was not able to reproduce when testing bug #1596595.