speckle-errors-transport

speckle-errors-transport

Diagnostic Decision Tree

Transport operation failed
|
+-- Network/HTTP error?
|   +-- 401 / 403 → AUTH EXPIRY (Section 1)
|   +-- 404 → WRONG STREAM ID (Section 2)
|   +-- 408 / timeout → TIMEOUT (Section 3)
|   +-- 502 / 503 / 504 → SERVER UNAVAILABLE (Section 4)
|   +-- ConnectionError / SSLError → NETWORK (Section 5)
|
+-- SQLite error?
|   +-- "database is locked" → SQLITE LOCK (Section 6)
|   +-- "disk I/O error" → DISK FULL (Section 7)
|   +-- "unable to open database" → PATH / PERMISSION (Section 8)
|
+-- Memory error?
|   +-- MemoryError / OutOfMemoryException → MEMORY LIMIT (Section 9)
|
+-- SpeckleException?
|   +-- "No transport" → NO TRANSPORT CONFIG (Section 10)
|   +-- "get_object not implemented" → WRONG METHOD (Section 11)
|
+-- Multi-transport failure? → PARTIAL FAILURE (Section 12)
+-- OperationCanceledException → CANCELLATION (Section 13)
+-- Progress callback throws → PROGRESS ERROR (Section 14)

---

Section 1: Authentication Expiry During Transfer

Symptom: HTTP 401/403 partway through a send/receive. Operation starts but fails after minutes.

Cause: Speckle tokens expire. ServerTransport stores the token at construction and reuses it for ALL requests. Long transfers can outlive the token.

Fix: ALWAYS create a fresh ServerTransport immediately before each operation. NEVER reuse across delays of more than 30 minutes.

# Create transport JUST before use
transport = ServerTransport(stream_id="abc123", client=client)
root_id = operations.send(base=my_object, transports=[transport])

// Set generous timeout for large models
var transport = new ServerTransport(account, "streamId", timeoutSeconds: 300);

---

Section 2: Wrong Stream ID (404)

Symptom: HTTP 404 on upload/download. Auth succeeds but operations fail.

Cause: The stream_id does not exist on the target server. Common when mixing IDs from different servers or after stream deletion.

Fix: ALWAYS verify stream existence before creating transport.

stream = client.stream.get(stream_id="abc123")  # Raises if not found
transport = ServerTransport(stream_id="abc123", client=client)

Rule: NEVER mix stream IDs from one server with credentials for another.

---

Section 3: Timeout Errors

Symptom: TimeoutError, requests.exceptions.Timeout, or TaskCanceledException. Common with models >100MB.

Cause: HTTP request exceeded configured timeout. C# default is 60 seconds per request.

Fix:

• C#: Set timeoutSeconds: 300 (or higher) at transport construction
• Python: Configure at session level or use client-based auth

Rule: ALWAYS set timeoutSeconds to at least 300 for models larger than 50MB. NEVER use the default 60-second timeout for production workloads.

---

Section 4: Server Unavailable (502/503/504)

Symptom: Intermittent HTTP 502/503/504 during batch uploads.

Cause: Speckle Server or reverse proxy is overloaded, restarting, or down.

Fix: Implement retry with exponential backoff. ALWAYS create a fresh ServerTransport per retry. NEVER retry immediately — wait at least 5 seconds.

for attempt in range(3):
    try:
        transport = ServerTransport(stream_id=sid, client=client)
        return operations.send(base=obj, transports=[transport])
    except Exception:
        if attempt == 2: raise
        time.sleep(2 ** attempt * 5)  # 5s, 10s, 20s

---

Section 5: Network Connection Errors

Symptom: ConnectionError, SSLError, ConnectionRefusedError, HttpRequestException.

Cause: No connectivity, DNS failure, SSL cert issues (self-hosted), firewall blocking port 443, or missing proxy config.

Fix: Verify connectivity before long operations. Configure proxy via HTTPS_PROXY environment variable. NEVER ignore SSL errors in production.

---

Section 6: SQLite Database Lock

Symptom: sqlite3.OperationalError: database is locked or SqliteException: SQLite Error 5.

Cause: SQLite uses file-level locking. Multiple processes writing to the same Data.db simultaneously causes lock errors. The default cache is shared across ALL connectors.

Fix: Use separate scope names for concurrent operations:

transport_a = SQLiteTransport(scope="ProjectA")  # Creates ProjectA.db
transport_b = SQLiteTransport(scope="ProjectB")  # Creates ProjectB.db

Rules:

• NEVER share the same scope across concurrent processes
• NEVER open Speckle SQLite with an external tool while connectors are running
• ALWAYS call close() (Python) or Dispose() (C#) when done

---

Section 7: Disk Full / I/O Errors

Symptom: sqlite3.OperationalError: disk I/O error or SQLite Error 13: database or disk is full.

Cause: The SQLite cache grows monotonically — Speckle NEVER purges cached objects. Heavy users accumulate 10GB+ caches.

Fix: Delete the cache file when no connectors are running (safe — recreated automatically). Use use_default_cache=False in serverless environments.

Rules:

• NEVER delete the cache while Speckle connectors are running
• ALWAYS monitor disk space on production machines

---

Section 8: SQLite Path / Permission Errors

Symptom: unable to open database file at transport construction.

Cause: Base path does not exist, no write permissions, or invalid characters. On Linux, $XDG_DATA_HOME may point to non-existent directory.

Fix: Specify a writable path explicitly:

import os
os.makedirs("/tmp/speckle-cache", exist_ok=True)
transport = SQLiteTransport(base_path="/tmp/speckle-cache", scope="MyProject")

Rule: NEVER assume %APPDATA% or ~/.config is writable in CI/CD — specify a custom base_path.

---

Section 9: Memory Exhaustion

Symptom: MemoryError or OutOfMemoryException during serialization or get_all_objects().

Cause:

1. Calling get_all_objects() on a large cache (loads everything into memory)
2. Serializing very deep/wide trees without detachment
3. Using MemoryTransport for large datasets

Rules:

• NEVER call get_all_objects() on databases larger than 100MB
• NEVER use MemoryTransport for datasets exceeding available RAM
• ALWAYS use @detach_property for large child collections
• ALWAYS set use_default_cache=True during send to enable streaming serialization

---

Section 10: No Transport Configured

Symptom: SpeckleException: Provide at least one transport...

Cause: send() called with transports=None AND use_default_cache=False.

Fix: ALWAYS provide at least one transport, OR leave use_default_cache=True (the default).

---

Section 11: Wrong Transport Method Call

Symptom: SpeckleException from server_transport.get_object(id) in Python.

Cause: Python ServerTransport.get_object() is NOT implemented — ALWAYS raises. Use operations.receive() or copy_object_and_children() instead.

# WRONG: server_transport.get_object(obj_id)  # RAISES
# CORRECT:
obj = operations.receive(obj_id, remote_transport=server_transport)

Rule: NEVER call get_object() directly on a Python ServerTransport. ALWAYS read from a local transport after copying.

---

Section 12: Multi-Transport Partial Failure

Symptom: Send completes but data is missing from one transport.

Cause: A failure in one transport does NOT abort the others. In Python, an exception propagates and may leave other transports incomplete.

Fix: Send to each transport independently for critical workflows:

for transport in transports:
    try:
        operations.send(base=obj, transports=[transport])
    except Exception as e:
        logger.error(f"Failed on {transport.name}: {e}")

Rule: NEVER assume all transports succeeded without verification.

---

Section 13: Cancellation Errors

Symptom: OperationCanceledException (C#) mid-transfer.

Cause: User or code cancelled via CancellationToken. Partial data may remain in transports.

Fix: ALWAYS pass a CancellationToken in C# interactive apps. NEVER attempt rollback — partial content-addressed data is harmless.

var cts = new CancellationTokenSource();
cts.CancelAfter(TimeSpan.FromMinutes(30));
await Operations.Send(data, new[] { transport }, cancellationToken: cts.Token);

---

Section 14: Progress Reporting Errors

Symptom: Exception from progress callback, not from the transport. Cross-thread UI errors in C#.

Cause: IProgress<ProgressArgs> callback invoked from a background thread. Updating UI without marshalling throws.

Fix: ALWAYS use new Progress<ProgressArgs>(callback) which captures SynchronizationContext. NEVER implement IProgress<T> manually without thread marshalling.

---

Retry Strategy Summary

Error Type	Retry?	Max	Backoff
Auth expiry (401/403)	Yes	1	Re-auth, new transport
Wrong stream (404)	No	—	Fix stream ID
Timeout (408)	Yes	3	Exponential
Server down (5xx)	Yes	3	Exponential (5s base)
Network error	Yes	3	Exponential (2s base)
SQLite lock	Yes	5	Linear (0.5s increments)
Disk full	No	—	Free space
Path/permission	No	—	Fix config
Memory exhaustion	No	—	Reduce payload
No transport	No	—	Fix config
Wrong method	No	—	Use correct API
Multi-transport	Yes	2	Retry failed only
Cancellation	No	—	User-initiated
Progress callback	No	—	Fix callback

---

Default Cache Paths

OS	Path
Windows	%APPDATA%\Speckle\Data.db
macOS	~/.config/Speckle/Data.db
Linux	$XDG_DATA_HOME/Speckle/Data.db or ~/.local/share/Speckle/Data.db

---

Critical Rules Summary

1. NEVER reuse ServerTransport across operations separated by >30 minutes
2. NEVER call get_object() on Python ServerTransport
3. NEVER call get_all_objects() on large databases
4. NEVER share SQLite scope across concurrent processes
5. NEVER delete cache while connectors are running
6. ALWAYS create fresh ServerTransport per operation
7. ALWAYS use separate scope names for concurrent SQLite access
8. ALWAYS pass CancellationToken in C# interactive applications
9. ALWAYS implement retry with exponential backoff for server errors
10. ALWAYS close/dispose transports when done

---

Reference Links

• references/methods.md — Transport API signatures and exception types
• references/examples.md — Working error handling code examples
• references/anti-patterns.md — What NOT to do with transports

Official Sources

• https://speckle.guide/dev/transports.html
• https://speckle.guide/dev/python.html
• https://speckle.guide/dev/dotnet.html
• https://github.com/specklesystems/specklepy
• https://github.com/specklesystems/speckle-sharp-sdk