3 ms·

GCP has had multiple many-hour (6+) GLOBAL outages in the past year. I think it's at about 3 so far this year. But, it doesn't make the headlines like a 2-hour

GCP has had multiple many-hour (6+) GLOBAL outages in the past year. I think it's at about 3 so far this year. But, it doesn't make the headlines like a 2-hour S3 outage in a single region, which must mean something ...

jpatokal 8y ago
GCP's status history would seem to disagree, unless you have unusual definitions of "outage" and/or "global": https://status.cloud.google.com/summary https://status.cloud.google.com/summary The last incident I'd personally classify as major lasted 39 minutes and was widely reported: https://status.cloud.google.com/incident/cloud-networking/18012 https://status.cloud.google.com/incident/cloud-networking/18... Disclaimer: I work at GCP but am not speaking for them. I also wish a speedy recovery for our colleagues at Azure: an outage like this can only result from many things going sideways simultaneously, and both the cause and recovery can be complicated in ways that flippant "well why didn't you just N+1 it" commenters here on HN can only guess at.
- ti_ranger 8y ago
  To be clear, by "global"(around the whole globe/earth), I mean that a specific service is broken in every public region . You seem to have a wider interpretation for "global". Besides the one you listed: * https://status.cloud.google.com/incident/compute/18005 https://status.cloud.google.com/incident/compute/18005 "Google Compute Engine VM instances allocated with duplicate internal IP addresses, stopped instances networking are not coming up when started." - 22 hours Newly-launched instances, or instances that were stopped and started, received duplicate IP addresses. 4.5 hours in a mitigation was provided, but it was only resolved after 22 hours, and customers may have had to still fix individual instances. As far as I remember, this was global, and there is nothing on the status page indicating it was limited to one region or a subset of regions. So, for 4 hours, if you needed to create a VM with working networking, you couldn't, anywhere on GCP, and no mitigation was available. Do you not consider this to be "global"? * https://status.cloud.google.com/incident/compute/18009 https://status.cloud.google.com/incident/compute/18009 "Instances using Local SSD might experience VM failures. This affects GCE VMs globally. No data corruption has been observed." - 5 hours The original claim was: > So AWS has had some big outages, as has Azure. Has GCP had any big outages yet? I said: > GCP has had multiple many-hour (6+) GLOBAL outages in the past year. I think it's at about 3 so far this year. So, maybe it's only 2 major global outages, or maybe it's 3 5-hour+ global problems, but the only way anyone can claim that Google hasn't had any big outages is if they don't have enough market share for a global outage to affect many websites or end-users.

jpatokal 8y ago

GCP's status history would seem to disagree, unless you have unusual definitions of "outage" and/or "global": https://status.cloud.google.com/summary https://status.cloud.google.com/summary The last incident I'd personally classify as major lasted 39 minutes and was widely reported: https://status.cloud.google.com/incident/cloud-networking/18012 https://status.cloud.google.com/incident/cloud-networking/18... Disclaimer: I work at GCP but am not speaking for them. I also wish a speedy recovery for our colleagues at Azure: an outage like this can only result from many things going sideways simultaneously, and both the cause and recovery can be complicated in ways that flippant "well why didn't you just N+1 it" commenters here on HN can only guess at.

ti_ranger 8y ago
To be clear, by "global"(around the whole globe/earth), I mean that a specific service is broken in every public region . You seem to have a wider interpretation for "global". Besides the one you listed: * https://status.cloud.google.com/incident/compute/18005 https://status.cloud.google.com/incident/compute/18005 "Google Compute Engine VM instances allocated with duplicate internal IP addresses, stopped instances networking are not coming up when started." - 22 hours Newly-launched instances, or instances that were stopped and started, received duplicate IP addresses. 4.5 hours in a mitigation was provided, but it was only resolved after 22 hours, and customers may have had to still fix individual instances. As far as I remember, this was global, and there is nothing on the status page indicating it was limited to one region or a subset of regions. So, for 4 hours, if you needed to create a VM with working networking, you couldn't, anywhere on GCP, and no mitigation was available. Do you not consider this to be "global"? * https://status.cloud.google.com/incident/compute/18009 https://status.cloud.google.com/incident/compute/18009 "Instances using Local SSD might experience VM failures. This affects GCE VMs globally. No data corruption has been observed." - 5 hours The original claim was: > So AWS has had some big outages, as has Azure. Has GCP had any big outages yet? I said: > GCP has had multiple many-hour (6+) GLOBAL outages in the past year. I think it's at about 3 so far this year. So, maybe it's only 2 major global outages, or maybe it's 3 5-hour+ global problems, but the only way anyone can claim that Google hasn't had any big outages is if they don't have enough market share for a global outage to affect many websites or end-users.