mirror of
https://github.com/hanzoai/migration.git
synced 2026-10-30 18:27:53 +00:00
zeekay
aa543d6d7d
The DOKS apiserver exec/cp stream intermittently truncates large transfers (unexpected EOF) — kms (884MB→59MB gz) failed mid-stream where hanzo_cloud (818MB gz) happened to succeed. This dumps+gzips ENTIRELY in-pod, splits into 8MB chunks, and pulls each with a per-chunk md5-verified retry loop, then verifies the whole-file md5 before accepting. gzip -t already catches truncation on the streaming path (pg-export.sh), so this is the fallback for any DB that fails there.
56 lines
2.7 KiB
Bash
Executable File
56 lines
2.7 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# pg-fetch-db.sh — robustly export ONE Postgres DB out of a flaky kube-exec path.
|
|
#
|
|
# The DOKS apiserver exec/cp stream truncates on transfers >~25MB (unexpected
|
|
# EOF), so a single `pg_dump | gzip` stream or one big `kubectl cp` loses data.
|
|
# This dumps + gzips ENTIRELY inside the pod, splits into small chunks, and pulls
|
|
# each chunk with a MD5-verified retry loop — so a dropped stream just re-pulls
|
|
# that chunk. Reassembles + verifies the whole-file md5 before declaring success.
|
|
#
|
|
# Usage: pg-fetch-db.sh <db>
|
|
# Env: NS(hanzo) POD(sql-0) DEST(~/Desktop/hanzo-pg-export) CHUNK(8m) TRIES(10)
|
|
set -uo pipefail
|
|
DB="${1:?usage: pg-fetch-db.sh <db>}"
|
|
export KUBECONFIG="${KUBECONFIG:-$HOME/.kube/config}"
|
|
NS="${NS:-hanzo}"; POD="${POD:-sql-0}"
|
|
DEST="${DEST:-$HOME/Desktop/hanzo-pg-export}"; CHUNK="${CHUNK:-8m}"; TRIES="${TRIES:-10}"
|
|
WORK="$DEST/.$DB-work"; mkdir -p "$WORK"
|
|
|
|
echo "→ [$DB] dump+gzip+split inside $NS/$POD (chunk=$CHUNK)"
|
|
kubectl -n "$NS" exec "$POD" -- sh -c \
|
|
'set -e; PGPASSWORD="$POSTGRES_PASSWORD" pg_dump -U "$POSTGRES_USER" --no-owner --no-privileges '"$DB"' | gzip > /tmp/'"$DB"'.sql.gz
|
|
split -b '"$CHUNK"' /tmp/'"$DB"'.sql.gz /tmp/'"$DB"'.part.
|
|
cd /tmp && md5sum '"$DB"'.sql.gz '"$DB"'.part.*' > "$WORK/manifest.txt" 2>/dev/null
|
|
|
|
if [ ! -s "$WORK/manifest.txt" ]; then echo " !! dump/split failed"; exit 1; fi
|
|
FULL_MD5="$(/usr/bin/grep -E " ${DB}\.sql\.gz$" "$WORK/manifest.txt" | /usr/bin/awk '{print $1}')"
|
|
echo " full md5 (in-pod): $FULL_MD5"
|
|
|
|
# Pull each chunk with md5-verified retry.
|
|
FAIL=0
|
|
while read -r want name; do
|
|
case "$name" in *.part.*) ;; *) continue ;; esac
|
|
got=""
|
|
for t in $(seq 1 "$TRIES"); do
|
|
kubectl cp "$NS/$POD:/tmp/$name" "$WORK/$name" >/dev/null 2>&1 || true
|
|
got="$(/usr/bin/md5sum "$WORK/$name" 2>/dev/null | /usr/bin/awk '{print $1}')"
|
|
[ "$got" = "$want" ] && break
|
|
sleep 2
|
|
done
|
|
if [ "$got" = "$want" ]; then echo " ✓ $name"; else echo " ✗ $name (gave up after $TRIES)"; FAIL=1; fi
|
|
done < "$WORK/manifest.txt"
|
|
|
|
if [ "$FAIL" -ne 0 ]; then echo " !! some chunk never verified — NOT writing $DB.sql.gz"; exit 1; fi
|
|
|
|
# Reassemble in split order + verify whole-file md5.
|
|
cat $(/usr/bin/ls -1 "$WORK/$DB".part.* | /usr/bin/sort) > "$DEST/$DB.sql.gz"
|
|
LOCAL_MD5="$(/usr/bin/md5sum "$DEST/$DB.sql.gz" | /usr/bin/awk '{print $1}')"
|
|
kubectl -n "$NS" exec "$POD" -- sh -c "rm -f /tmp/$DB.sql.gz /tmp/$DB.part.*" >/dev/null 2>&1 || true
|
|
rm -rf "$WORK"
|
|
|
|
if [ "$LOCAL_MD5" = "$FULL_MD5" ] && gzip -t "$DEST/$DB.sql.gz" 2>/dev/null; then
|
|
echo "✓ [$DB] complete + md5-verified — $(/usr/bin/du -h "$DEST/$DB.sql.gz" | /usr/bin/awk '{print $1}')"
|
|
else
|
|
echo "✗ [$DB] reassembly md5 mismatch ($LOCAL_MD5 vs $FULL_MD5)"; exit 1
|
|
fi
|