Saturday, June 20, 2015¶
Magics with Python fixtures¶
I am working on an exciting problem of data loss on a production site. This problem took me already 20 hours.
195 partner records (of 15446) “vanished” from the database without letting any trace. Unfortunately we started to discover the problem only more than a week after it occurred, and furthermore we did not have any daily snapshots. The last known good backup was from 20150605. So simply restoring a backup was not an option.
This is a situation where Python dumps are a cool thing.
By the way: the cause of the problem is unknown so far. A human mistake in one of the two management tasks that had been done during this week? A pitfall in Lino (e.g. the fact that Lino accepts setting a circular reference on countries.Place.parent)? A technical problem on the server? We have had about five prime suspects so far, but each of them turned out to be innocent after some research. We don’t even know for sure whether they disappeared all at once or in several series. Though it seems that “it” happened on 20150611.
Anyway, fortunately the customer is now on vacation for three days, so I have a long weekend for analyzing and repairing their data.
Basic strategy is as follows: find out the IDs of the lost clients and inject these from the 20150605 snapshot.
Some clients had been re-created by reading their eID card again. This caused the following error messages after injecting the vanished back into the database:
INFO Deferred Client #200715 (u' (200715)') : (1062, "Duplicate entry '710608 576-38' for key 'national_id'")
INFO Deferred Client #200717 (u' (200717)') : (1062, "Duplicate entry '720619 037-74' for key 'national_id'")
INFO Deferred Client #200559 (u' (200559)') : (1062, "Duplicate entry '801112 221-12' for key 'national_id'")
INFO Deferred Client #200708 (u' (200708)') : (1062, "Duplicate entry '580515 498-14' for key 'national_id'")
INFO Deferred Client #200716 (u' (200716)') : (1062, "Duplicate entry '800721 476-41' for key 'national_id'")
INFO Deferred Client #200714 (u' (200714)') : (1062, "Duplicate entry '950120 575-51' for key 'national_id'")
INFO Deferred Client #21148 (u' (21148)') : (1062, "Duplicate entry '710108 503-76' for key 'national_id'")
By searching for the SSIN in 20150605/pcsw_client.py
I could
determine which were the original partner id of these clients:
200708 '580515 498-14' 4011
200714 '950120 575-51' 20737
200715 '710608 576-38' 20606
200716 '800721 476-41' 22564
200717 '720619 037-74' 20450
200559 '801112 221-12' 23093
The last message had a different reason. It seems that the national_id of client 21148 has been modified to become that of client 6655:
21148 '710108 503-76' 6655
This modification is currently the prime suspect of being the cause of our problem. But to reproduce it, I’d need to write a test case in a MySQL database (not SQLite as usual), which might take some time. It is more probable that partner 6655 had simply been deleted before the modification.
Related objects which had been created for these clients need to get attributed to the real partner id. Here is an example of the warnings which visualized these problems:
- dupable_clients.Word {'owner': [u'Client instance with pk 21148 does not exist.']} (2 object(s) with primary key 14, 15)
- dupable_clients.Word {'owner': [u'Client instance with pk 200717 does not exist.']} (2 object(s) with primary key 16944, 16943)
- dupable_clients.Word {'owner': [u'Client instance with pk 200714 does not exist.']} (2 object(s) with primary key 16941, 16942)
- dupable_clients.Word {'owner': [u'Client instance with pk 200559 does not exist.']} (2 object(s) with primary key 16934, 16933)
- aids.RefundConfirmation {'doctor': [u'Person instance with pk 902 does not exist.']} (4 object(s) with primary key 17, 167, 301, 330)
- pcsw.ClientContact {'client': [u'Client instance with pk 21148 does not exist.']} (1 object(s) with primary key 1738)
Here is a snippet of code inserted into restore.py
in order to
fix these duplicate clients:
DUPS = {
200708: 4011, # '580515 498-14'
200714: 20737, # '950120 575-51'
200715: 20606, # '710608 576-38'
200716: 22564, # '800721 476-41'
200717: 20450, # '720619 037-74'
200559: 23093, # '801112 221-12'
}
def new_partner_id(old):
return DUPS.get(old, old)
And the following functions needed a manual additional line of code to call new_partner_id:
create_aids_granting
create_aids_incomeconfirmation
create_notes_note
create_dupable_clients_word
create_cal_guest
create_cal_event
create_pcsw_coaching
create_pcsw_clientcontact
There was also again a series of warnings for a yet unidentified problem (looks as if users sometimes change the date range of grantings after having issued a confirmation):
aids.IncomeConfirmation [u’Date range 1/1/15… lies outside of granted period 1/1/15…4/21/15.’] (1 object(s) with primary key 144)
aids.IncomeConfirmation [u’Date range 6/1/15…6/1/15 lies outside of granted period 12/1/14…5/31/15.’] (1 object(s) with primary key 383)
aids.IncomeConfirmation [u’Date range 1/1/02… lies outside of granted period 1/1/02…5/31/15.’] (1 object(s) with primary key 248)
aids.IncomeConfirmation [u’Date range 12/1/14… lies outside of granted period 12/1/14…5/31/15.’] (4 object(s) with primary key 125, 126, 127, 129)
Finally, the following warnings were normal and confirm that our duplicate clients have been removed):
contacts.Partner {u’id’: [u’Partner with this ID already exists.’]} (6 object(s) with primary key 20450, 20606, 22564, 20737, 4011, 23093)
pcsw.Client (1062, “Duplicate entry ‘20737’ for key ‘PRIMARY’”) (1 object(s) with primary key 20737)
pcsw.Client (1062, “Duplicate entry ‘23093’ for key ‘PRIMARY’”) (1 object(s) with primary key 23093)
pcsw.Client (1062, “Duplicate entry ‘20450’ for key ‘PRIMARY’”) (1 object(s) with primary key 20450)
pcsw.Client (1062, “Duplicate entry ‘22564’ for key ‘PRIMARY’”) (1 object(s) with primary key 22564)
pcsw.Client (1062, “Duplicate entry ‘20606’ for key ‘PRIMARY’”) (1 object(s) with primary key 20606)
pcsw.Client (1062, “Duplicate entry ‘4011’ for key ‘PRIMARY’”) (1 object(s) with primary key 4011)
Making a snapshot of a Lino database¶
I worked two more hours on setting up a system for making auotomatic daily snapshots of a Lino database. Also started a documentation page about it: Making a snapshot of a Lino database.