Smrtovrisk / google-refine

Automatically exported from code.google.com/p/google-refine
Other
0 stars 0 forks source link

Fetch URLs aborts while saving/loading/computing facets #440

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago
What steps will reproduce the problem?
1. Run a Fetch URL for 30 mins.
2. GETS are found in log with saving workspaces (no problems so far)
3. After about 30 mins, a loaded project happens where /core/get and 
/core/compute operations fire and Fetches stop running.
4. Auto Saving continues to fire and run every 5 mins during the entire 
sequence and after the Fetches stop running.

--
Notice the timing of all of this... on WithActiveFacetOn7000Rows.txt

a quick save, then a load from disk with compute-facets
(perhaps since I did have a facet open with 7000 rows filtered True?
because I was trying to Fetch URLs only on those filtered 7000 rows and not all 
my rows)
then after the compute-facets the whole Fetch/GET process just stopped.
Maybe we need an Auto-Save checkmark option in Preferences ?
Looking at 127.0.0.1:3333/preferences only allows adding a key value.  Do we 
have have defined for AutoSave = False ?
Maybe it's a bigger bug and Auto Saving could/can work with Fetching URLs and I 
just am unaware ?

I tried another Fetch URLs run with NoFacets_Fetching_All_8800_Rows.txt , and 
the problem occured again, with similar timing events after about 30 mins as 
well.

(Also sent David H. my private MySpace.tsv file to help debug this.)

Original issue reported on code.google.com by thadguidry on 25 Aug 2011 at 5:03

Attachments:

GoogleCodeExporter commented 9 years ago
I believe this is being triggered by the cache purge (which is done as part of 
the autosave processing).  Autosave is 5 min. by default and main memory is 
purged of any projects which haven't been modified in the last hour.  The 
modification time isn't updated until the operation completes and is posted to 
the history, so any long running operation has the potential to have its 
Project object (which it's in the process of using) deleted from memory in the 
middle of the operation.

I've applied a fix in r2222 which I think should cure the problem (not only for 
this, but all long running operations).  Please give it a try and let me know 
how you make out.

Original comment by tfmorris on 28 Aug 2011 at 10:02

GoogleCodeExporter commented 9 years ago

Original comment by tfmorris on 28 Aug 2011 at 10:10

GoogleCodeExporter commented 9 years ago
So, does this turn off ANY writes to disk, until a Fetch URLs operation 
completes ?  If not, what types of writes to disk would still happen, even 
after this patch (logging, anything else) ?  Asking so I can test further, 
later on.

Original comment by thadguidry on 29 Aug 2011 at 1:18

GoogleCodeExporter commented 9 years ago
Pulled new code via SVN, cleaned, build, ran Fetch URLs operation on 2400 
MySpace urls, which ran for about 1.5 hours (unattended) but I did watch it up 
to 90%, walked away, when I returned to look at the Ubuntu terminal (20 mins 
later), I saw the Fetch1 column created, but no data for any rows?  I quickly 
captured the Log (attached here).  I then did a Shift F5 refresh, and no data 
in column showed.  I stopped Refine and restarted with same parameters and 
opened Firefox again, still no data showed in Fetch1 and a isBlanks facet 
showed True on all 2400 rows.  (sigh) Back to the drawing board? 

Original comment by thadguidry on 29 Aug 2011 at 5:17

Attachments:

GoogleCodeExporter commented 9 years ago
I could have sworn a replied to comment 3, but it looks like it never posted.  
Writes aren't affected at all by the change, the only thing that has changed is 
that projects with long running processes aren't purged from memory.

Having said that, I don't think the project is marked as modified until the 
operation completes and it won't get autosaved unless its modified.

re: comment 4 were there any errors in the UI?  You mentioned something about 
using facets before.  Was this test done with a facet or facets active?  If so, 
can you try  a smaller test with and without facets to see if that changes the 
behavior?  I'll look to see if there is a path where the operation can 
terminate without making the error visible.

Original comment by tfmorris on 30 Aug 2011 at 8:02

GoogleCodeExporter commented 9 years ago
No facets used on the Comment 4 testing.  I am strictly testing the Fetch 
operation to reach a 100% completion and also see the "CellAtRow 
fetch(CellAtRow urlData)" at line 253 of 
ColumnAdditionByFetchingURLsOperation.java actually create "new Cell" in the 
column.  It definitely creates the column on 100% completion... however it 
doesn't appear to be inserting the stream into each new Cell...for some weird 
reason.  That is the general failure that I am seeing.

I would categorize this as 2 issues, I think?

1. Fetch doesn't always reach 100% completion, No error shown, No new column of 
data.
2. Fetch DOES reach 100% completion, DOES Create a new column, however no 
stream data appears in the column's new cells (such as raw HTML).

Original comment by thadguidry on 30 Aug 2011 at 8:25

GoogleCodeExporter commented 9 years ago
I have tried 2 more rounds of testing with Fetch URLs.
1. Fetch 2000 rows of unique URIs from "test A" domain. (about 2 hours)
2. Fetch 4500 rows of unique URIs from "test A" domain. (about 3 hours)

Both tests completed and full HTML content was added into a new column.
Both tests used 2500ms delay. (quite harsh 2.5 sec expectation on returning a 
page result)
Both tests used multiple facets to filter down to those particular rows.

My conclusion is that this seems now fixed, and that my previous concern seems 
isolated to some sort of MySpace problem, or perhaps just too aggressive with 
the delay during testing against the MySpace domain.  The different "test A" 
domain is also well known, and exhibited no such problems at all with using 
current Trunk r2234 (after Tom and David's patches).

We may want to leave open until after the 2.5 public beta testing. And if no 
further issues, then close.

Original comment by thadguidry on 7 Sep 2011 at 6:35

GoogleCodeExporter commented 9 years ago
Based on Thad's additional testing and my analysis of the problem, I'm pretty 
convinced this is fixed.

Original comment by tfmorris on 22 Sep 2011 at 10:34

GoogleCodeExporter commented 9 years ago

Original comment by dfhu...@google.com on 9 Oct 2011 at 5:22