We are using the Google Homegraph API to implement REQUEST_SYNC functionality for our Google Home project. This has appeared te be working fine in our testing, but recently we noticed that roughly a third of requests are responded to with HTTP codes other than 200.
We have seen the following codes:
400 Bad Request
500 Internal Server Error
429 Too Many Requests
404 Not Found
429 is most likely due to some fault on our part, and we are currently looking into avoiding this. However, we are unable to determine what is causing the other codes.
In our limited ability to test, we have found that it appears to be limited to a specific user each time. Unlinking and re-linking our service seems to fix the issue (if only temporarily).
As a side note, the example here appears to contain invalid JSON, as there are quotation marks missing around agent_user_id. When testing a simple request, the omission or inclusion of the quotation marks didn't affect its success, but this may be confusing to readers.
Does anyone have advice to offer regarding these issues?
EDIT:
We are currently trying to reproduce the issues for an account by purposefully sending many REQUEST_SYNCs. This causes a number of 429 responses (and we've seen a single 500), but it has not yet caused future REQUEST_SYNCs to stop working entirely - previously we saw a user only ever receive 500 response codes until the service was relinked.
Unlinking a user did cause subsequent REQUEST_SYNCs to respond with 404, as expected. Is there a way to be notified when a user is no longer using our Google Home service so that we may stop sending them SYNC requests when they change something about their devices?
A 400 may happen if there's an issue with your server sending data, such as if your agent_user_id has not been established in the Home Graph but you're trying to call something like REQUEST_SYNC.
A 500 may happen if the data you're returning has an issue, like if it doesn't match the schema or is otherwise not formatted correctly.
A 429 may happen if you're hitting the API too often.
A 404 may happen if the URL you're requesting doesn't exist, such as if the API key is invalid or the parameters you're sending are invalid.
If you can identify reproducible cases, it would be easier to find a fix.
Related
Since few days the adaptive payments (both preapprovals and chained) seems to be broken.
When we send the user to paypal to complete the order (chained) or to authorize the preapproval (preapprovals), they get a generic "invalid transaction" error. Tested the flow, we get the same error and looking to the console we can see this error on the checkout.js file in the paypal object directory:
Uncaught TypeError: PAYPAL.AP.Subflow is not a constructor
at checkout.js:11
(anonymous) # checkout.js:11
Seems something has broken on paypal side.
Anyone with the same issue? Workarounds?
Cheers, Fil
We've finally figured out what was wrong.
Trying the complete debugging of the call to the svcs endpoint we've discovered that since few days (first time from 2013), we started getting an odd response from the paypal server to our SOAP request as:
{"faultstring":"Unknown Content-Encoding","faultcode":"HTTP"}
We've then thought to some issue in the "transportation" of the call to paypal servers and we've tryed to add something in the headers.
First, we discovered that in our http header request some line that we've never used before was missing, for example:
"X-PAYPAL-DEVICE-IPADDRESS: " . $_SERVER['REMOTE_ADDR'] . "\r\n" ;
But this didn't solve the issue.
Then we've added a new line in the soap client array:
'compression' => SOAP_COMPRESSION_ACCEPT,
Adding this solved the issue but no one ever adviced us about the requirement of this part of the code. Probably, this is a linux issue, maybe paypal has changed some server with a newer one and while the older accepted the SOAP request even if it was not with this "compression" parameter, in the newer environment it seems required.
Anyway, everything is back to normality adding this line, we just still have an odd time response (long) for the Pay operation while everyting is smooth for the Preapproval operation but this might depend on different paypal servers managing one or the other operation.
We use the https://financialdatafeed.platform.intuit.com/v1/accounts/account_id_goes_here/transactions endpoint on a recurring basis to fetch transactions for all of the accounts we sync. We've been using this stably for quite awhile now, across a wide variety of accounts spanning 100s of financial institutions. This works great.
However, occasionally we get a report from a user who claims that we're not receiving transactions that they know to exist. Our investigation protocol is as follows:
To ask the user if they see the transactions when they sign into their bank's web site directly
To ask them to confirm that the credentials they used on their bank's web site are precisely the ones that they entered when setting up credit card sync on our site
We then manually inspect the response body from the above mentioned URL, to make sure that the HTTPS response indicates HTTP 200 and has a non-error response body (our app catches these errors correctly, but if debugging mysteriously missing transactions, we inspect the response body visually).
We look to see whether we're successfully syncing transactions for any other user that relies on the same FI. If we are, we become confident that both the bank and Intuit APIs are well-behaved, and that the problem is on our end somehow.
We sometimes ask users to try the same FI in Mint, guessing that if it fails in Mint, that it might be a bank or FI issue.
Investigation steps 1-2-3-4-5 tease out the root cause of at least 99% of the times when a user emails us to say that we're not successfully receiving their transactions. However, the remaining 1% are the tricky ones.
Today I'm faced with a situation where a user sees the txns on their bank website, swears that they are using the same creds when adding the card to our site, the HTTP response from the endpoint is HTTP 200 but contains zero transactions, but yet when the user tries via Mint they successfully see transactions.
However, the particular FI (OnPoint Community Credit Union) is not one where I can do investigation step 4, because we have no other users that currently rely on that FI. Is it possible for someone at Intuit to check to see whether there is evidence that users relying on OnPoint Community Credit Union are currently, successfully, retrieving transactions from that particular FI?
Any other suggestions for how to further deduce whether the zero-transaction response is due to: (a) user error, (b) bank server responding incorrectly, (c) Intuit server responding incorrectly, vs (d) our app behaving incorrectly?
Can you please submit a support ticket to Intuit with the Account_ID that is missing the transactions so that we can diagnose the issue? The first place to start when diagnosing the issue is to look at the Agg_status_code to make sure that reflects a '0'. If we are unable to login due to invalid credentials or MFA might be a cause of the missing transactions. I can help diagnose though once a ticket is submitted.
Regarding this API: https://developers.google.com/admin-sdk/email-audit/#accessing_account_information
I have been using the Admin SDK to retrieve login history for users in our Google Apps for Business setup. When I request individual users at a time, the request sometimes takes a few hours to process (in which the state is PENDING). However, when those few hours pass, I still get the login history that I need.
The problem continues as I begin requesting more users. We have around 750 users, and of those 750~ requests I made, 725 gave me an error after waiting ONE WEEK for my requests to be processed. Even worse, the ones that did not error out are still pending! Here is the response I get when I check the status of a request that errored out:
{'status': 'ERROR', 'adminEmailAddress': '***#etsy.com', 'requestDate': '***', 'requestId': '***', 'userEmailAddress': '***#etsy.com'}
This has got to be the flakiest and most unreliable API I have ever been unfortunate enough to work with. Requests can take anywhere from an hour to over a week to process, with no indicator of success in the mean time. Errors can also happen for no apparent reason, and no messages or explanations as to why.
It looks like this issue has been resolved by the Google Engineers. Try to run the calls again. It shouldn't be in pending more than the "normal" expected time. I just tried earlier, and I was able to export login info for my users.
After paypal updated their interface (sandbox.paypal.com for example is not working, now you have to go to developer.paypal.com) many of the things are not working: 2 of them are particularly frustrating and I was hoping someone here knew how to get around them:
Am I the only one whose sandbox customer test accounts are not able to make purchases? The transaction page says they are not available.
IPN validation is not letting me send a https request. When I do it says there is something wrong with the server name. Yesterday however before the update I could get verified status. If I dont put https, now my handler gives me an invalid responde status, code: 400. What does it mean?
To fix the HTTP 400 error, follow the instructions in https://www.x.com/content/bulletin-ipn-and-pdt-scripts-and-http-1-1 and update your code to pass "Host" information. Ideally, things should work with just the recommended changes from the above link. Apparently, thats not the case. Here is a fix from one of the PayPal MTS person - PalPAL sandbox IPN processor rejecting all messages?
Remove the "cmd=notify-validate" option from the validation URL. I tried this and it worked. Though it doesn't return the right string, atleast it doesnt break with the 400 error.
While we wait for a fix from Paypal, I wonder how a company like PayPal can cause such a huge blunder and not post anything on their status page - https://www.x.com/developers/paypal/documentation-tools/site-status/pp-cri. It just makes you think that even smaller companies can do a better job than companies like PayPal.
For the code:400 issue, you have to update the post to version 1.1. That information is located here.
https://www.x.com/content/bulletin-ipn-and-pdt-scripts-and-http-1-1 in this bulletin.
However, as I posted before the asp.net example uses a call, that does not exist, so I was only able to get mine partly working. After fixing this, the servers appear to be rejecting calls to https, or the cert they have installed is invalid.
Action Required before February 1, 2013
Merchants need to update their IPN and/or PDT scripts to use HTTP 1.1, and include the “Host” header in the IPN postback script. In addition to this bulletin, these merchants will be notified via a direct email.
Alright, seems to be fixed!
If you are having trouble logging in, like suggested above, clear cache and cookies and try again.
Regarding the error 400, seems to have been solved by paypal!
With the Graph API, I publish a story by POSTing to the /me/feed connection. I get back a success or an error result from Facebook. So far so good. Once in a while, the API takes a long time and the connection times out. In that case, I don't know for sure if the request succeeded of failed (i.e. maybe the request never reached Facebook, or maybe it succeeded and the result never made it back to me). How do you handle this situation?
More details:
I publish a lot of posts to Facebook and Twitter, so the timeout situation happens often. With Twitter, the solution is easy. If the request times out the first time, I simply try again. Twitter detects duplicates, so if the post was successfully published the first time, then I'll get a "duplicate status" error on the second request and I know that I don't need to retry any more.
But Facebook doesn't detect duplicates, so if I retry the publish request, I risk having two copies of the post published to the user wall, which is not nice. On the other hand, if I don't retry, I risk having the post not published at all. Thoughts?
I get back a success or an error result from Facebook.
Hmmm. When I post to the Graph API, I get back an error or the id of the post. I never see any success message. What SDK are you using around the API?
Once in a while, the API takes a long time and the connection times
out.
Usually when things are running slowly, it's due to the channelUrl not being specified. See https://developers.facebook.com/docs/reference/javascript/
It is important for the channel file to be cached for as long as
possible. When serving this file, you must send valid Expires headers
with a long expiration period. This will ensure the channel file is
cached by the browser which is important for a smooth user experience.
Without proper caching, cross domain communication will become very
slow and users will suffer a severely degraded experience.