Gcp Notes

Auth

get the auth file and then:

export GOOGLE_APPLICATION_CREDENTIALS="/usr/home/user/.gcp/XXX-XXX.json"

whoami ?

gcloud auth list

gsutil version -l

The gsutil will show legacy boto files:

${HOME}/.config/gcloud/legacy_credentials/david.thornton@domain.com/.boto

but in the same dir there is:

${HOME}/.config/gcloud/legacy_credentials/david.thornton@domain.com/adc.json

which you can put in the GOOGLE_APPLICATION_CREDENTIALS env var.

There a couple of env vars, it's not clear when to use which one. It's a bit all over the place. At this time GOOGLE_APPLICATION_CREDENTIALS works in the most places I care about ( terraform )

echo ${GOOGLE_CREDENTIALS}
echo ${GOOGLE_CLOUD_KEYFILE_JSON}
echo ${GCLOUD_KEYFILE_JSON}

Projects

A logical place to put your stuff.

Use this "bag" as a billing unit.

In as much as you want to use label for billing, some charges can't be labels. Project partition that cost.

list your projects:

gcloud projects list

Not all project have billing accounts.

Labeling and filters and formatting at the command line

https://cloud.google.com/blog/products/gcp/filtering-and-formatting-fun-with

Storage

Types of storage, how to choose:

https://cloud.google.com/storage-options/

Outputs

extra one label for a coloumn in a table out put:

gcloud pubsub topics list --format="table[box](name:sort=1,labels[product])"

┌───────────────────────────────────────────┬──────────┐
│                        NAME               │ PRODUCT  │
├───────────────────────────────────────────┼──────────┤
│ projects/($project)/topics/(topic name )  │ (product)│
│ ...                                       │ ...      │
└───────────────────────────────────────────┴──────────┘

Formating Output

Here is an example of the way in which things in the api returned response can be formatted for display in a table:

(gcloud.artifacts.repositories.list) Expected ) in projection expression [    table[title="ARTIFACT_REGISTRY"](
         name.basename():label=REPOSITORY,
         format:label=FORMAT,
         mode.basename(undefined=STANDARD_REPOSITORY):label=MODE,
         description:label=DESCRIPTION,
         name.segment(3):label=LOCATION,
         labels.list():label=LABELS,
         kmsKeyName.yesno(yes='Customer-managed key', no='Google-managed key'):label=ENCRYPTION,
         createTime.date(tz=LOCAL),
         updateTime.date(tz=LOCAL),
         sizeBytes.size(zero='0',precision=3,units_out=M):label="SIZE (MB)"
    ) table[](name,cleanupPolicyDryRun,sizeBytes *HERE* /1000000000)].

Compute

https://cloud.google.com/sdk/gcloud/reference/compute/instances/create

how do I like project and familiy for well known images for terraform builds?

gcloud compute images list  --standard-images

list non-running instances

gcloud compute instances list

_always_ show _only_ running.

But what about the failed, initializing, terminated instances? Try this:

gcloud compute instances list --filter="status:*"

on scaling:

https://cloud.google.com/compute/docs/autoscaler/understanding-autoscaler-decisions

( Log still don't tell you on what metric is used to decide to scale. )

List instances lit up with a template

gcloud compute instances list --filter="metadata.items.list(show="keys"):instance-template"

or list name and template:

gcloud compute instances list --format='value[](name,metadata.items.instance-template)'

list template with metadata item X

gcloud compute instance-templates list \
 --format='value[](name,properties.metadata.items.X)'

Port forwarding via gcloud

connect to the locally running port:

gcloud compute ssh --zone us-central1-c <<instancename>> --verbosity=info -- -NL 4545:localhost:4545

now connect to localhost:4545 and get the cloud instance.

OS Login

So you want to just ssh into the vm like you do everything else, you don't want to use

gcloud compute ssh ...

or the "in browser" ssh client.

great , you want "OS Login"

lots of steps:

1. for the VM set the enable-oslogin meta data value to "TRUE"

in tf like this:

  metadata  = {
    enable-oslogin = "TRUE"
  }

2. give the user the correct roles:

Computer OS Login ( for vanilla , non-root access )
Compute OS Admin login ( for root access via sudo )

via command line this I think:

gcloud projects add-iam-policy-binding project-ID --member \
 serviceAccount:"USERNAME@project-ID.iam.gserviceaccount.com" \
 --role "roles/iam.serviceAccountUser"
 --no-user-output-enabled --quiet

loop over all hosts

get a list of names and zones:

gcloud compute instances list --filter="labels.label_name=label_value" --format="table[no-heading](name,zone)" > ~/tmp/instancelist

compute login needs the vm's zone.

while read name zone
do
echo gcloud compute ssh --zone \"$zone\" \"$name\"  --project \"${project}\" --command=\"important command\"
done < ~/tmp/instancelist > ~/tmp/out.sh

then sh out.sh

Start up script basic web sewrver

/basic-web-server-startup.sh

Netowrking

packet loss query:

fetch gce_zone_network_health :: networking.googleapis.com/cloud_netslo/active_probing/probe_count | 
map add[pd_local_zone: resource.zone, pd_remote_zone: metric.remote_zone, pd_local_region: resource.region, pd_remote_region: metric.remote_region] | 
filter (pd_local_region =~ 'us-central1') | 
filter (pd_remote_region =~ 'us-central1') | 
{
     filter metric.result = 'failure'     | 
     group_by [resource.zone, metric.remote_zone], 14400s, .sum     ;     group_by [resource.zone, metric.remote_zone], 14400s, .sum     | 
     filter val() >= 1 '1'
} |
ratio |
top 50, .mean

from "Network Intelligence" -> "Performance Dashboard" -> "Packet Loss"

How Tos

single node NFS

https://medium.com/google-cloud/gke-with-google-cloud-single-node-filer-nfs-4c4dc569964f

Annoyances

1. 'gcloud compute ssh' requires that you know what zone the instance is in.

2. compute instances name change is destructive. Change the name? destroy and recreate :(

3. Web gui search bar "redis" doesn't return "memorystore" ( gcp branded redis ) - has been FIXED as of 2021

4. Web gui search bar "memorystore" yields no results. - has been FIXED as of 2021

5. Web gui search bar quite slow. - has been FIXED as of 2021

6. Web gui Load Balancers: default view not good for professionals. you must always go to "advanced" ( small text at the bottom of the list ). No way to configure it to always go to advanced.

7. Web GUI load balancer, hard to "see" internal load balancers.

8. Instance log by their ID. You cannot search by instance name in logs. To map a name to an ID search this:

resource.type="gce_instance"
protoPayload.request.name="INSTANCE_NAME"
protoPayload.methodName="v1.compute.instances.insert"

Then look in the protoPayload.response.id field. Then change your search to:

resource.type="gce_instance"
protoPayload.response.id="INSTANCEID"

sigh.

Gotchas

scripting

you are making a GetClusterRequest object in your python ... but api docs , which version are you using?

https://cloud.google.com/python/docs/reference/dataproc/latest/google.cloud.dataproc_v1.types.GetClusterRequest

don't do this:

request = container_v1.GetClusterRequest(
  project_id=myproject,
  region=myregion,
  cluster_id=mycluster
)

do this:

request = container_v1.GetClusterRequest(
  name=f"projects/{myproject}/locations/{region}/clusters/{mycluster}"
)

trraform intenral lb exanple

https://cloud.google.com/load-balancing/docs/l7-internal/int-https-lb-tf-examples

Logging

Queries of note

-resource.type="k8s_cluster"
-resource.type="http_load_balancer"
-resource.type="vpn_gateway"
-resource.type="gce_instance"
-resource.type="cloudsql_database"
-resource.type="pubsub_topic"
-resource.type="pubsub_subscription"
-resource.type="stackdriver_notification_channel"
-resource.type="audited_resource"
resource.type="gce_instance_group"

are you over quota somewhre?

There is a quota dashbarod, but what about the log:

protoPayload.status.message="QUOTA_EXCEEDED"

Load Balancer Request logging

Which of your back end have logging turned on, and what is their sample rate?

gcloud compute backend-services list --format json | \
jq '.[] | { name: .name , enable: .logConfig.enable , samplerate: .logConfig.sampleRate}| join(",")'

Getting logs with Python

Sometimes you just can't get what you want from the UI.

So do it yourself:

#!/usr/bin/env python3

# use:
# 'gcloud logging logs list'
# to list availble logger_names.

import json
from google.cloud import logging
import pprint

logger_name    = "compute.googleapis.com%2Fhealthchecks"
logging_client = logging.Client()
logger         = logging_client.logger(logger_name)

for entry in logger.list_entries():
  timestamp = entry.timestamp.isoformat()
  # uncomment this to figure out the fiels you want.
  # pprint.pprint(entry)
  print("{} {}".format( entry.payload['healthCheckProbeResult']['probeSourceIp'],entry.payload['healthCheckProbeResult']['healthState']))

In this case I could not get log metrics.

Reference: https://googleapis.dev/python/logging/latest/entries.html

Log Metrics

Load balancer requests

In this example we are asking:

Which grafana dashboards are being used the most?

First setup a logging metric for l7 load balancer, then:

fetch l7_lb_rule
| metric 'logging.googleapis.com/user/lblogmetrics'
| filter
    (resource.forwarding_rule_name
     == 'k8s-fws-cluster-monitoring-grafana--XXX')
    &&
    (metric.requesturl
     =~ 'https://grafana.domain.com/d/.*')
| align rate(1m)
| every 1m
| group_by [re_extract(metric.requesturl, "^https://grafana.domain.com/d/(.*)\\?", r'\1')],
    [value_lblogmetrics_aggregate: aggregate(value.lblogmetrics)]

Here I've relabels "re_extract..." as "url"

fetch l7_lb_rule
| metric 'logging.googleapis.com/user/lblogmetrics'
| filter
    (resource.forwarding_rule_name
     == 'k8s-fws-cluster-monitoring-grafana--XXX')
    &&
    (metric.requesturl
     =~ 'https://grafana.domain.com/d/.*')
| align rate(1m)
| every 1m
| group_by [ url: re_extract(metric.requesturl, "^https://grafana.domain.com/d/(.*)\\?", r'\1')],
    [value_lblogmetrics_aggregate: aggregate(value.lblogmetrics)]

Note the re_extract function in the group by, for group just the part of the request you care about. some request urls can get quite long, such that they don't fit in graphs well.

You can do this sort of slice and dice for any site in GCP.

Be careful with logging. For very busy sites google logging might be expensive, you can also do logging ratios to get a feel for activity without logging _every_ request.

Monitoring and Alerting

Overquota events

https://cloud.google.com/monitoring/alerts/using-quota-metrics

Find orphaned instances

i.e. instances not owned by a mig are orphans

#!/bin/sh

# A. list instance groups
# B. list all instance in instance group
# C. list all instance
# D. substract one from the other.

echo A. gcloud compute instance-groups list

export INSTANCE_GROUPS=$(mktemp)
echo $INSTANCE_GROUPS
gcloud compute instance-groups list | grep region | grep -v NAME | awk '{print $1}' > ${INSTANCE_GROUPS}

echo B. gcloud compute instance-groups managed list-instances

export NOT_ORPHANS=$(mktemp)
for i in `cat ${INSTANCE_GROUPS}`
do
 gcloud compute instance-groups managed list-instances $i --region us-central1 | grep -v NAME >> ${NOT_ORPHANS}
done

echo C. list all instance

ALL_INSTANCES=$(mktemp)

gcloud compute instances list | awk '{print $1}' > ${ALL_INSTANCES}
for i in `cat ${ALL_INSTANCES}`
do
  grep -q $i ${NOT_ORPHANS} || echo orphan $i
done

gke instance might show up in this.

Survey

VMs

Redis - "memorystore"

SQL

Also see

/gcp cloudevents

Reading

Hashes and ETags

Best Practices

https://cloud.google.com/storage/docs/hashes-etags