github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2025-07-27 21:47:07 +00:00
Code Issues Projects Releases Wiki Activity

Merge pull request #67378 from mborsz/log-dump

Browse Source 
Automatic merge from submit-queue (batch tested with PRs 67378, 67675, 67654). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.

Store logs from 'logexporter' to allow debugging it.

**What this PR does / why we need it**:
With https://github.com/kubernetes/kubernetes/pull/67190 logexporter stopped working properly in 5000 kubemark test (while it works fine in smaller tests)
As we have no tools to debug this in a big scale, I propose storing logexporter logs in some place so that it's possible to debug all attempts.

**Special notes for your reviewer**:

**Release note**:

```release-note
NONE
```
This commit is contained in:
Kubernetes Submit Queue 2018-08-22 02:45:05 -07:00 committed by GitHub
commit 3084408ac9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 31 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
cluster/log-dump/log-dump.sh
View File

@ -54,7 +54,7 @@ readonly systemd_services="kubelet kubelet-monitor kube-container-runtime-monito
# Limit the number of concurrent node connections so that we don't run out of # Limit the number of concurrent node connections so that we don't run out of
# file descriptors for large clusters. # file descriptors for large clusters.
readonly max_scp_processes=25 readonly max_dump_processes=25
# TODO: Get rid of all the sourcing of bash dependencies eventually. # TODO: Get rid of all the sourcing of bash dependencies eventually.
function setup() { function setup() {
@ -197,23 +197,23 @@ function dump_masters() {
return return
fi fi
proc=${max_scp_processes} proc=${max_dump_processes}
for master_name in "${master_names[@]}"; do for master_name in "${master_names[@]}"; do
master_dir="${report_dir}/${master_name}" master_dir="${report_dir}/${master_name}"
mkdir -p "${master_dir}" mkdir -p "${master_dir}"
save-logs "${master_name}" "${master_dir}" "${master_logfiles}" "" "true" & save-logs "${master_name}" "${master_dir}" "${master_logfiles}" "" "true" &
# We don't want to run more than ${max_scp_processes} at a time, so # We don't want to run more than ${max_dump_processes} at a time, so
# wait once we hit that many nodes. This isn't ideal, since one might # wait once we hit that many nodes. This isn't ideal, since one might
# take much longer than the others, but it should help. # take much longer than the others, but it should help.
proc=$((proc - 1)) proc=$((proc - 1))
if [[ proc -eq 0 ]]; then if [[ proc -eq 0 ]]; then
proc=${max_scp_processes} proc=${max_dump_processes}
wait wait
fi fi
done done
# Wait for any remaining processes. # Wait for any remaining processes.
if [[ proc -gt 0 && proc -lt ${max_scp_processes} ]]; then if [[ proc -gt 0 && proc -lt ${max_dump_processes} ]]; then
wait wait
fi fi
} }
@ -258,7 +258,7 @@ function dump_nodes() {
nodes_selected_for_logs=( "${node_names[@]}" ) nodes_selected_for_logs=( "${node_names[@]}" )
fi fi
proc=${max_scp_processes} proc=${max_dump_processes}
for node_name in "${nodes_selected_for_logs[@]}"; do for node_name in "${nodes_selected_for_logs[@]}"; do
node_dir="${report_dir}/${node_name}" node_dir="${report_dir}/${node_name}"
mkdir -p "${node_dir}" mkdir -p "${node_dir}"
@ -266,17 +266,17 @@ function dump_nodes() {
# many nodes. # many nodes.
save-logs "${node_name}" "${node_dir}" "${node_logfiles_all}" "${node_systemd_services}" & save-logs "${node_name}" "${node_dir}" "${node_logfiles_all}" "${node_systemd_services}" &
# We don't want to run more than ${max_scp_processes} at a time, so # We don't want to run more than ${max_dump_processes} at a time, so
# wait once we hit that many nodes. This isn't ideal, since one might # wait once we hit that many nodes. This isn't ideal, since one might
# take much longer than the others, but it should help. # take much longer than the others, but it should help.
proc=$((proc - 1)) proc=$((proc - 1))
if [[ proc -eq 0 ]]; then if [[ proc -eq 0 ]]; then
proc=${max_scp_processes} proc=${max_dump_processes}
wait wait
fi fi
done done
# Wait for any remaining processes. # Wait for any remaining processes.
if [[ proc -gt 0 && proc -lt ${max_scp_processes} ]]; then if [[ proc -gt 0 && proc -lt ${max_dump_processes} ]]; then
wait wait
fi fi
} }
@ -347,6 +347,28 @@ function dump_nodes_with_logexporter() {
sleep 15 sleep 15
done done
# Store logs from logexporter pods to allow debugging log exporting process
# itself.
proc=${max_dump_processes}
"${KUBECTL}" get pods -n "${logexporter_namespace}" -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.spec.nodeName}{"\n"}{end}' | while read pod node; do
echo "Fetching logs from ${pod} running on ${node}"
mkdir -p ${report_dir}/${node}
"${KUBECTL}" logs -n "${logexporter_namespace}" ${pod} > ${report_dir}/${node}/${pod}.log &
# We don't want to run more than ${max_dump_processes} at a time, so
# wait once we hit that many nodes. This isn't ideal, since one might
# take much longer than the others, but it should help.
proc=$((proc - 1))
if [[ proc -eq 0 ]]; then
proc=${max_dump_processes}
wait
fi
done
# Wait for any remaining processes.
if [[ proc -gt 0 && proc -lt ${max_dump_processes} ]]; then
wait
fi
# List registry of marker files (of nodes whose logexporter succeeded) from GCS. # List registry of marker files (of nodes whose logexporter succeeded) from GCS.
local nodes_succeeded local nodes_succeeded
for retry in {1..10}; do for retry in {1..10}; do